I deal with this all day long at work and it’s exhausting. People almost acting like no one has thought of it “I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?”
What kills me is you might expect this from a busy high level manager that doesn’t really understand the technical details and they just point the AI to an error they got. They don’t know how to interpret the response, so they ask someone who work on the thing. It’s still kinds annoying because you could just ask, but whatever. But to get these from junior and senior engineer for the areas they work in and expect someone else to read it for them? It’s crazy behavior. How can someone serious even think that’s ok.
At my dayjob there is a person spearheading ai across the enterprise.
They generated lots of documentation across the whole stack and now makes all PO/BAs read it if it's correct. So not just 300 lines - he unironically generated thousands of lines of "documentation" and is now making hundreds of people review it for him
Complete brainrot
Au psychosis is getting seriously outrageous at this point
Thankfully I'm a dev and thus aren't in the blast radius of that genius idea
I think a lot of that is people who were always very insecure because they’re mediocre engineers. Previously asking questions or not understanding something was a bit painful but now you can hear “you’re absolutely right!” and make progress all day every day.
Until you need to interact with actual humans and that’s why you try to minimise it, hiding behind ai generated content.
Everyone I know, including myself, is simultaneously extremely pro-AI and anti-AI for this reason.
I always put myself in the shoes of the reader when relaying Claudeslop. I either use it as a reference and write it by hand or make Claude edit it 10+ times until it's conveying only the needed information, and in simple, clear words, and without annoying Claudeisms.
How correct is the documentation? You seem to be coming from the same standpoint as a lot of people in this thread which is "don't make me read AI output, I can prompt the model myself" but what this person is doing sounds much more sane - have highly skilled subject matter experts prompt the AI once, generate all the answers you might want, and review them carefully for correctness using all available context.
That sounds far preferable to individuals independently asking questions that are not covered by the documentation and independently reviewing a bunch of potentially hallucinated slop that no one else will ever read.
I am that person on the blast radius. The most effective response I have found is dumping the documents into the AI, and asking it to review them, then posting back the AI response.
Any resistance to this is met with a "But I'm also using AI as the board demands!"
It's okay, when enough terrible decisions impact a company they tend to do really poorly thought out layoffs so you'll be in the blast radius sooner or later.
On the contrary, I kind of love this as form of malicious compliance. I'd wager money that the PO/BAs have been demanding "comprehensive documentation" for a very long time, and now they are getting what they've asked for good and hard.
When some people call this "Verification Debt" or something similar, that is setting the Overton Window upon this phenomena adversely upon the verifiers. I call it an application of Brandolini's Law:
"The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.""
Until you put in at least the same amount of energy required to verify the AI output as you are asking others to put in to verify it, the AI output you are wielding must be considered bullshit. Really, it should be based upon the verification requestor's historical trust score and how many human verifiers are looped in. A Verification Tax, so to speak:
The energy the creator must spend (Ec) is determined by a function of the number of reviewers requested (N) and the historical trust score of that creator's AI outputs (T).
The thing I think is so frustrating about this is it's the absolute worst case middle ground between two much more productive experiences. Either solve the problem yourself by understanding the code or just give me the full agent context so I can just use a chatbot too.
For high stakes debugging efforts I'll dig in with Claude, have it do a bunch of testing and give me a writeup, and then review it to extract any useful findings. When discussing with other people, I give them the 2 sentences I'm confident in and then link them to the giant doc so they can review it with their Claude.
IMO once you're working with agents, your whole job is context management. I have a little web server running my Claude markdown docs. When someone sends me a Claude snippet with some partial information I just have them prompt their bot to upload a thorough context dump so I actually know what they did. If someone has had an agent look at a problem they haven't thought about yet, I'd rather just get the info from the bot directly.
what is often overlooked in corporate culture: if you take a step back, two coworkers refusing to cooperate, regardless of the reason, is a typical "corporate kindergarten" scenario for the manager to take care of.
To avoid rushed or incorrect decisions, conflicts intentionally move through gradual escalation. If someone behaves unprofessionally, for example by submitting a code review they do not understand, the first step is to establish the likely cause.
They may ultimately be at fault, but that does not absolve the surrounding environment. The culture may be toxic, deadlines unrealistic, or communication poor.
The best course of action for someone directly affected by uncooperative coworkers is to avoid assuming ill intent. Get them on a call and let them explain their pull request, however trivial the issue may seem. Even if the gesture is misinterpreted, you still have a far stronger position than righteous indignation.
Because engineers don’t decide on hiring and firing. The managers all have KPIs to transform the org into AI-native. This is the kind of behavior that organizations encourage right now.
I get it’s annoying and I personally hate it but firing someone for it is ridiculous. Are you expecting everyone to rewrite everything from an AI before they share it because that doesn’t sound a good use of company time either.
It seems that this is largely coming from higher ups who are probably (consciously or not) testing the waters for LLMs being able to replace the people they're dumping the walls of text on. I don't mean in any sort of malicious penny pinching scheme either, but simply people wanting to be able to do things quickly and as close to their own demands as possible, while getting rid of the 'middle man.' It just so happens that as LLMs become ever more capable of delivering on e.g. software tasks, software engineers start to become the middle men.
Because it's harder to fire managers. And even harder to fire executives.
Who goes when companies need to downsize? Rarely executives; they'll always find a way to be retained...hell, they're the ones with the power to decide who goes.
Let me give you a scenario and see if it is in this same fireable offense line:
An Ops person who deals with company code base but is not a SWE in the code base regularly has to work around issues with this software (They are OPs). However they have read access to the source.
In the past, they see a bug and open a jira with how it was found for the dev team, and it may or may not ever get looked at.
Now ops can point claude to the repo and it can most likely [but not always] find the issue.
So this ops person can do most of the work of reproducing the bug and finding the problem/solution but doesn't understand the code. Should they not produce the findings for someone else to read/validate? Should they go back and just open the ticket the old way?
I had an encounter with the opposite problem the other day, PM asked me if the library I was using for randomization was "secure" and I said, "Yes, I used the standard Java library for cryptographically secure random numbers that is already used in several other locations in this project," and he said, "But did you ask Claude if it's secure? It has to be secure!"
“I ask Claude only questions to which I don’t know answers to - which there are plenty of. That saves the company money by avoiding frivolous use of tokens.”
> and it spit out this 300 line response. Can you read it for me and see if it’s right?
If I get a request like this on a day when I don't feel sufficiently quixotic, I say sure and just put it through an LLM myself - if they don't bother to re-read it, why should I? This does not have a great promise for a future of work, of course. Also, in my teams, I explicitly tell people not to do that.
> But to get these from junior and senior engineer for the areas they work in and expect someone else to read it for them? It’s crazy behavior. How can someone serious even think that’s ok.
Deciphering error messages can be a bit of an arcane art, at times impenetrable to all except the authors of a system. Understanding the patterns exposed in such messages and, thereby, deriving their meaning is not always clear even to experienced developers.
Engineers reaching out to you (humans) to explain the output of an LLM haven't fully internalized how to interact with LLMs. When an LLM's output is beyond one's immediate grasp, one can _ask the LLM_ to further break down its output, to explain the concepts and symbols that inform its response.
Personally, I have asked LLMs to draw charts and comprehensively simplify their output, often providing cognitive signposts that "I'm confused" about some aspect of what its reporting.
In other words, some people, including engineers with various lavels of experience, don't yet fully understand the implications of the open-eneded interface that language provides with LLM interaction.
i learned working in customer support that if you give someone a helpline their brain stop working. they will ask any little stupid thing even if in reality they can know the answer, look it up or reason about it. they are not dumb but the helpline cripples them...
AI is a helpline, so for a lot of people it cripples them. they cease to think. simple as that.
its something people need to be aware of that their brains work like this, because that awareness is the same thing that solves the problem. (people find it very confronting to learn they have been asking questions they knew the answers too already...)
the other day I had an "official" career conversation with my boss. They took some notes during the talk and later sent me a page summary of the conversation to see if it looked right. It immediately smelled of AI, and had some things completely opposite to what I said. jeeze I guess this is where we are headed...
I've literally seen this in HN responses as well, so many times. What the hell is the point of human interaction if the "human" just forwards a chat bot's response
At least those (at least the detectable ones) seem to be downvoted en masse, called out, or flagged. They aren't even allowed as per the guidelines[0]:
> Don't post generated text or AI-edited text. HN is for conversation between humans.
Exactly! When someone does it at work, I get a bit sad because my conclusion is that whoever does it can't like their job very much. But that's ok – not liking your job very much is compatible with an otherwise happy life.
When people start doing it in an entirely voluntary human-to-human situation, like having a conversation online about a topic of interest, I'm gobsmacked.
You can't just remember every conversation and ... something useful.
Indiscriminate memory buries you. What you need is "monumental" memories, which is also known as documentation. AI can be very effective in helping you manage that. That should be kept tight, just like code. You want to keep enough of a record of how you got to where you are that you are equipped to be effective in future tasks.
I would say that your suggestion is basically digital hoarding. Lets say there are 3 kinds of people, the person with OCD who keeps his house almost barren in the name of tidiness, the person who visits his possessions periodically with a critical eye, deciding what to keep and what to throw away, and the hoarder. When it comes time to solve some household problem, say hang a picture, which one of them will have what they need to do it (a hammer and a nail, maybe a little bit of wire) and be able to find it?
I think you might consider keeping the session history for a while, maybe a month or 3 months or 6 months, but have something that automatically deletes them after that.
I'm not opposed to straight copypasta, but it should be limited to excerpts only or at the very least workshopped with the LLM to reduce verbosity. Certainly it should at least be fully read and grokked.
I can count on one hand the number of times I've seen LLMs one-shot responses that are fit to paste to Slack verbatim. It does happen though.
I think it was reasonable in the very recent past. There was a period where AI had major capabilities that had not diffused fully into the zeitgeist. So often you could see someone struggling with e.g. understanding a crash log, and say "they probably haven't thought of pasting into Claude Code with access to the codebase".
It’s truly unhinged behavior that has me questioning how much longer I want to remain in the industry. You always knew that 80% of the people on any team or in any large company just didn’t care. But now you see who these people are out in the open.
I had a guy with more senior title say “I love the new api and cli. Is anyone working on the docs? Because I can take a crack at them” then goes to generate 12 markdown files, and shares them with me. All the cli commands and APIs are wrong. I was very confused. You have the api spec, and the cli. You couldn’t point your agent to them to validated the commands and the api? He replies “oh that’s a good idea. I just described them to the agent and thought you could review and fix mistakes”
This one kills me. So many forums, and Reddit in particular, now have people just chiming into topics with ChatGPT answers. It gives them the illusion of being helpful, while they don't really have the knowledge of the subject. But people go to communities to converse with others in the field, and this makes it feel dead internet.
I'm glad most of my colleagues are still doing things themselves. But honestly, if all you can do is tell me what Claude told you and you didn't even bother to read it yourself. Then I might as well just cut you off the team and go use Claude directly.
Eh, you always had to filter the chaff too before. Nowadays there's just a second group that replaced knowledge pulled from one's bottom with knowledge pulled from ChatGPT, if you weren't aware of the former group at least now you're forced to be on guard due to the latter.
You know what's even worse? When they don't even bother to copy-paste. They just have claude code/codex post a slack message, with all the telltale signs of an AI message (caution emojis everywhere!)
> People almost acting like no one has thought of it
Everyone at my job uses LLMs to code, all day long. People still ask me "why does this bug happen" and I use Claude to find out and the answer is almost always straightforward, and when its not I raise a discussion with the team. The calculus isn't "can I prompt an LLM to find out," it's "do I have a solid enough mental model of the part of the codebase to prompt an LLM specifically enough, and also understand if the LLM is hallucinating or not."
> How can someone serious even think that’s ok.
I think this is a culture issue. I think every org is different, but we had a similar ramp up at our company. First we trialed cursor. Then some people were committing slop. Then we were like "llms are no excuse for slop." Then we got our skills and context good enough to not require close review. And then everything became LLM driven and now we have become "meat proxies" where everyone asks LLMs to fix things that are part of unfamiliar parts of the codebase and then asks someone else familiar to review the fix. We're tending towards a solution to this because I dont think its a good usage of resources but every org will come up with their own culture and solution to this.
When was the last time that a newly adopted technology revealed that amount of ... potential for improvement in our fellow humans? AI must have really hit some nerve to generate this massive amount of questionable behavior. Or did I fall prey to bias?
It's so real.. Suddenly managers sending MR, commenting MRs, all AI driven.. 300 lines text, where you don't even know if the premises are correct. I guess untrained people with better tools are super dangerous. I don't know how to stop them..
I’ve been using V/G as shorthand for this: verification capacity / generation capacity.
LLMs have made G extremely cheap, but verification capacity and judgement has not scaled at anything like the same rate. Someone can generate a bunch of PRs or a report quickly, while transferring hours of verification work to everyone downstream.
We are on the wrong side of this ratio, because we never had the capacity to extinguish reviewer capacity the way we can today.
“Don’t be a meat proxy” needs to be a cultural norm for the larger problem of using cheap generation to externalise verification costs onto someone else.
We had the same problem, now I just ask my colleagues to AR the docs (attentionrequest.com); while the site may be tongue-in-cheek, it helped us build a workflow, where "certificate of having read & thought about it" is conditioned on reading the doc and then writing up a number of questions on it as a human, before sending it off. It's worked reasonably well for us so far, hopefully this becomes one of the new norms.
The fact that we’re starting to point these people out gives me great comfort.
I believe in proper breaks, more than enough downtime, work-life balance - but I’ve been hamstrung my entire career by people who seem to be so low-bandwidth that they often seem to be operating on autopilot - while making the same (or far more) as I’m making.
This is because they don't trust you to do it. If whenever a problem came up you and the rest of the team were able to solve it and it didn't drag on into some forever bug, they wouldn't be second guessing you. Maybe they have long experience where you were too lazy to fix things or you bamboozled them with status updates that were basically fabricated? Don't act offended, it's quite common.
I created a spreadsheet called “slop” and add the name to the list and redirect all queries to their engineering manager. I’m not wasting my time on that shit.
I will refuse to read such messages or documents. At the very minimum, i want a human-written TLDR-style summary to ensure the sender has read their own slop.
Feels like some serious hyperbole. I definitely have people at work copy and paste Claude output but I have never once had someone ask me to read it for them and tell them it was correct.
I think this is just the moment we are in, the amount of people using AI in their day to day work changed so quickly that we have a very wide spread range of competencies and experience and really nobody has had enough time to be an expert with AI. We don't have well defined best practices and it feels like the target is moving.
My guess is you were probably complaining about something similar to this before there was AI, this is just the new target for your angst.
It's implicit as a request, not something they literally say (usually).
Namely - I put near zero effort into this response, and deciding if it's correct is on you. -> I'm asking you to read it and figure out if it's correct because I couldn't be bothered to do so
At my $work, people do this frequently, but instead of it just being insensitive, they have their personal Slackbot responding on their behalf. After I realized I wasn't even talking to a person, I've stopped opening their DMs.
One way to prevent obvious AI language from sneaking in text destined for other human beings is to ask the model to produce ASD-STE100 Simplified Technical English bullet points.
This will result in a list of sentences that are clear and explanatory, easier to double-check, and convenient for the user to rewrite into a more readable format with a human voice.
I've been thinking about how to avoid having AI-isms creep into my own voice, because I see that way too much in my tech-world friends. It's natural enough: if you spend a lot of time talking to somebody, you usually end up mirroring some of their speech patterns.
One idea I had was to start doing my Claude sessions in German. (That happens to be my strongest second language, no other reason.) Thus any Claude-isms I pick up will presumably not wander into my English, which is what I use for most communications.
This would have the added benefits of getting some much-needed German typing practice, and preventing me from ever lazily copy-pasting AI text into an email or a README.
+1 on this being a great idea. I also need to learn how to write in German (I can speak at B2 and read C1, but I find writing difficult as I've never bothered tbh). The added benefit of avoiding Claudisms is interesting. I think I'll join you on this.
With the recent wave of posts about Claude rewriting entire codebases from Zig/Electron to Rust/Swift, I can't help but question the prompter's actual confidence in the final result. Gaining a thorough understanding of the generated output still demands an enormous amount of effort. I suspect the reality is that most people just chain a few review agents together, and if they output "LGTM", the code gets committed and pushed.
I have been doing daily agentic coding myself since last November, but lately Im having concerns. We are on the verge of becoming Warhammer 40k Tech-Priests: chanting sacred prompts to appease the machine spirit, praying for a good result. The foundational knowledge is lost, meaning that even if we do look at the code, we won't understand it anyway.
> The foundational knowledge is lost, meaning that even if we do look at the code, we won't understand it anyway.
but this is similar to a lot of other technical fields that have width and depth. People would not understand assembly any more and blindly trust the output of the compiler. People now can only design a part of a CPU, and not the entirety. Chip manufacturing is so complicated that there's probably no single person who could understand the entire process, from end to end.
Most developers by any metric haven't hopped onto the full agentic AI bandwagon. People who want to stop understanding code will indeed largely lose their technical skills, but they're a minority of developers. Its one of the reasons why I think its a terrible idea to invest in personally, being skilled at something has never once been a negative for me in my entire life even if it requires work
Not yet. The world still pretty much runs on systems not written by LLMs. And humans wrote all of that. LLMs runs on OSes and hardware created by "not LLMS".
And seen the "quality" of the output of LLMs (and I pay three AI subscriptions), I don't think we're getting rid of human-produced code, OSes, hardware, anytime soon.
It's interesting to me, I have never once felt this. I have basically just been excited by using AI for coding. I've been a software engineer for 15 years and before that a physics PH.D. student (and 1 year professional scientist) who focused largely on computational work. I love programming and I've felt reinvigorated by the introduction of AI and developing systems around it.
I tend to think it's just different personality types. There are a lot of different type in engineering. There's a significant set of people that really love to just be given detailed, well described set of work and crank it out, they don't really want to be involved in much outside of just implementing the plan. That is not all engineers but certainly a chunk of them. I could see that group feeling a bit sidelined by AI. But that's not me, so I am just speculating.
I enjoy and use AI a lot both for personal projects, but also at work. I think that feeling of “we’ve become claude code babysitters” is a lot more prevalent in larger teams and/or in work environments where … most people have resorted to claude coding all the time.
You can’t not claude code if everyone else is pushing to main (or opening prs which get ai reviewed), and once your days become mostly prompting and ai code review, you start remembering the days (mere months ago) where work did not feel or look like this at all.
The condom image is both an exaggeration, but is also almost a too-accurate depiction of what work has become - very rapidly - for many people.
I’m sure people were also very excited to get their first cars after years of riding on horses. It doesn’t mean new kinds of pollution (and accidents) are not happening, or that we don’t miss horse riding haha
We once interviewed a candidate for a senior engineering position and when asked what his primary role was at his current job, he said "I mostly make sure that nobody changes the software".
He wasn't wrong, and his role was still valuable to the company.
The mental model I like to use is something along the lines of the LLM is a firehose and my brain is a mere straw and that we have to eventually get all the important parts to trickle through my straw sized brain. Or another way, the LLM is all the dough shoved into a pasta extruder and all the dough needs to eventually extrude out of my brain into prod-pasta.
At my last job, a coworker did this to me. The first time it happened, I ignored it. The second time, I responded in public saying “thanks but I can ask Claude myself.” Nobody ever pasted me an LLM response again. YMMV with team size and seniority though
At my current job I just asked a coworker in chat if the reply was his or LLM. He quickly owned up and the incidence of brokered LLM messages in chat dropped signficantly.
Unfortunately people are being encouraged to use LLM in docs, emails, and presentations so it's probably a losing battle.
We have a LLM bot in chat that analyzes bugs when.you drop in a link and anything else if you ask it directly. It's actually quite effective: all the slop is now localized and can be ignored/blocked easily, and people no longer feel the need to copy and paste slop themselves.
I've been thinking about these kind of interactions a lot lately. I don't know what it says about human communication patterns but I'm not super excited about it.
I've been guilty of acting as a middleman for review comments from difficult coworkers. I farm out my interaction to avoid the confrontation, mental exhaustion, and to avoid negative feelings I got from those reviews. I don't mean 'my work is being challenged', I mean people who: delay, nitpick, blow up architecture after not participating in design, assert false info as truth and require you prove them wrong, require 10x evidence of everyone else. Not people who are like 'Hey, I don't understand this code can you add some comments or lets hop in a room for review'
I've also been on the receiving end. I like to think I'm NOT one of these toxic reviewers.
Now, we have Business Owners and Managers who are now even further away from engineers. Important meetings have 0 engineers in them ( a couple people who think they're engineers because they were for 5 years in the 80s ). Output is an AI generated transcript that gets copy/pasted into copilot to generate jira tasks by the Product Owner.
It seems like we (humans) really don't like working with other people. Or, maybe we like working with people who can advance our careers and want to automate interaction with people who cant?
I think this effort by Business Owners to set up an AI wall may be the closest I've seen to 'digital classism' (maybe there is a better term for it?).
I think if you work somewhere that this is the result, you're unlikely to have much progress challenging anything directly. In order to actually have change you need to be able to challenge and stamp out these sorts of behaviours.
This smells an awful like a post-hoc justification of why you can’t answer it yourself.
If you _could_ do that and hand the information to Claude, you could also just…give the person that information directly, and avoid wasting their time and the tokens.
If I've asked *you* a question, it's because I want to know what *you* think.
I don't want to know what Claude thinks.
If you don't know the answer, say that. "I don't know" is a perfectly good answer, albeit one which I might follow up with something like "Well why don't you know? This is literally your job to know", which is kind of what 90% of my emails at work for the past three weeks have been.
If someone asked you a question, it's most likely because they value you. Whether it's your skills, your domain knowledge, or something else, people ask you questions because they know you can provide some value. When you give them a Claude response, not only are you being rude, but you're devaluing yourself in the eyes of those around you. They learn that you can't be relied upon any more, and so they stop asking you questions.
You took all this time to reformulate the question, add the relevant implicit context, and do back and forth BUT you did not have time to reformulate the answer to something human readable that YOU own and can take responsibility for? This does not make sense to me. Nobody said not to consult an llm while formulating an answer. But if you give an answer, own it. Don't throw slop grenades.
Arguably we already have, if you concentrate on intelligence rates since 1975: the reverse Flynn affect. Those who grew up about that time no longer had to generate their own entertainment through storytelling, make their own music, calculate things mentally: everyday they had music, entertainment pumped into their homes, a way to avoid athematic, etc. And now we're evening thinking a lot less in our jobs, not even practicing the skills we learnt. Intelligence is only going to decrease.
When Plato complains about writing diminishing our faculties for memorizing things, it's sometimes treated as an example of how we've always pointlessly worried about these things, but he was absolutely correct and the transition from an oral culture to a written culture has absolutely demolished those capabilities.
Writing as a technology ended up having strong upsides that mostly made up for that (in particular our ability to communicate across distances and time, to read Plato's concerns without 100 generations of Chinese whispers), but it was a trade-off, as most technologies tend to be.
It makes me so angry that an entire branch of "science" gets to just take data and go "I dunno man, maybe kids are just dumber". There's so much of this in social science. Take some measurable thing, and then just wildly speculate about the cause of it in an entirely untestable way, and then test that theory in ways that absolutely do not justify the conclusions. At the very least, the whole field should get a lesson in how to communicate the certainty of their theories and conclusions. You've got some guys hanging around couching their uncertainty when they're banging sub atomic particles together at the speed of light and then you've got some social science bro being like "If you move that checkbox on that form to the bottom it'll be way more effective, trust me bro I've hacked how people think".
It’s tempting to use what engineers were 70 years ago vs now as an analogy. From a technical point of view, I was told we would be struggling to solve the maths and physics exercises our grand fathers took as students in college, and they had to deal with additional disciplines that disappeared like industrial design (ie drawing blueprints by hand) or manual calculus. So it is tempting to think that the level of engineers has gone down.
But a modern engineer has computers, math software, can code, has all the knowledge of the internet at their fingertips. With all that assistance I believe modern engineers are way more productive and capable.
> With all that assistance I believe modern engineers are way more productive and capable.
That isn't what was asked. The capabilities of "human + machine" clearly are increasing, but the comments above posit that as a result the average capabilities of "human (sans machine)" may decrease. The relative value of each change is far from clear.
I work in the embedded space, and I routinely have to work with technologies I'm initially not familiar with. Internet searches are very hit and miss. So, is "AI". I generally use every tool at my disposal to try to get my foot in the door so I can understand better questions to ask and get the technical information I need to do my job.
I absolutely have no problem with someone replying to my question with a response from Claude. I appreciate them telling me the source, so I can assign an appropriate level of trust in the information.
Not exactly. A drawing program could just create a new "blank slate" that substitutes some old skill requirements for new skills. The key point is that SOME machines create a sufficient mass of "blank slate"-like environments for active creativity and others (like AI) only diminish that room for creatvity to flourish. The danger is creating and using machines of the latter type.
I think you are confusing machines with motors. Many machines do things that simply are not possible without. A bicycle, for example, is not at any less work than walking, but it enables you to go faster than you can run. Menually operated sewing machines produce stitches that would be infeasible by hand.
Motorising any of those machines could be considered lazy, though.
Human evolution does not occur in a vecuum. I suspect there will still be preasure to get smarter (how ever way you define that) because of competition with other humans.
Do you guys think we're going to see a de-evolution of human beings due to technology?
The gap between people who are willing to commit the hours to learn something and those who prefer the path of least resistance is getting bigger.
I’d like to add that for hardware development (in my experience) it’s been non-stop double checking because of hallucinations. To the point where I don’t use it for weeks.
What's even more worrying is that I've heared plenty of people defend it as, "Younger generations are just intelligent in different ways compared to older generations."
I'm halfway through my 30's and this downturn started with my generation, halfway through the 80's. I hope I'm missing something, because I still need to see a benefit for lower attention spans, lower comprehensive reading skills and lower critical thinking capabilities.
Not sure about de-evolution (devolution is the word I think), just a change in how people and brains work - but you could already see this happening the past few decades. I can't read a technical book or documentation end to end ever since google results came up with specific answers within seconds.
A technical book can be so relaxing (unless e.g. too highly technical... think quantum mechanics). Imagine 50k words on an interesting tech topic without a context switch!
I think we're going to see a sort of "intelligence caste" form. People who overuse AI and atrophy their brain's ability to think will be on the bottom, while the inverse of those people will be on top.
This is somewhat good news because it means the end (to some extent) of the Peter Principle. My hope, though, is that this doesn't turn into an actual caste system, but I'm not holding my breath. I think that's where this all goes: we're in the process of the world's intelligence being strip-mined and resold back to the world at a premium. The only ones who "survive" long-term are the ones who refuse to submit all of their mental faculties to AI.
That inevitably leads to a rapid degeneration of the species. Only way to avoid that outcome entirely is to remind people that their brain is valuable and they need to exercise it just like they would the rest of their body.
Well, apparently, our brain sizes have shrunk something like 30% over the last 10,000 years. This is primarily because ancient man needed to be a jack of all trades just to survive in the wilderness. Once civilization and specialization took over we just didn’t need all that processing power since a lot of that work was just outsourced to civilization and society.
Alcohol has existed for zillions of years, but not literally every person is a useless drunkard.
You could consider it a class of predator on people who have vulnerabilities that can give way to dependence on the LLM and an inability to function. It could be seen in that light. The underlying problem is being described in that light.
> Do you guys think we're going to see a de-evolution of human beings due to technology?
A de-evolution of culture for the sake of efficiency, sure. The entire culture will shift to a greater degree of superficiality, not just because of the lack of deep understanding of what they're doing, but because AI will condition (and is conditioning) people to want the more superficial stuff.
This isn't a phenomenon exclusive to AI - but AI is at the apex of it and is the worst we've seen so far because it, much more than anything else, targets the mental domain whereas most previous automations have targeted the physical but the mental as well to a lesser extent.
The end result is a huge amount of technical debt that simply can't be recovered and a huge increase in myopia that will prevent us from truly understanding and solving serious problems that will undoubtedly occur.
Most people can't see this because they're already addicted to the dopamine hit of the tech, but it's one reason I'm 100% anti-AI and I hate it with an immense passion.
I am lucky that I work with great teams and so when someone says "this is how Claude (or Gemini) summaries the situation" it does mean "I read it and it's right-enough to be helpful, so I am passing it along without editing"
Obviously for this to work you need (1) a team that's smart and mature enough to own the AI output it puts forward and (2) shared understanding that people are operating this way.
I do find that when you have 1 and 2 this is a real accelerant. I asked a sales rep the other day what was going on in a key account and he forwarded along a 10 page Claude synthesis.
The reality is this is better intel than I would have been able to get in a pre-AI era this quickly. And yes once I went into the weeds on the doc I found one thing that didn't make sense and I asked the rep he stared at it and said "you are right that's a hallucination I don't notice" but overall it was still worth it.
So the answer was "What's going on with this account" was 10 pages of homework, with at least 1 mistake to correct? That doesn't seem like an improvement over a conversation.
I suspect that we will end up with a social convention or understanding around reputation that is tied to whether someone pushes slop on other people. If I need to pass on some AI slop, retyping and copy-editing works well to validate whether it's reasonable while I work it over in my mind.
We already have social patterns for disregarding "Well... Acktchully" Bros who jump into conversations where they're not welcome and I think we can easily build up a new rule for "Well... Claude says 'Acktschully'" Bros
I agree. Also there's one specific context where I use this quite a bit, even if it's a bit passive aggressive...:
When someone is asking me about something that is a) their job and b) I have explained to them before. There are a few specific people that I have to do this to repeatedly because for some reason they don't want to do the thing their team owns. It's a high noise source at my work and this response is my new version of RTFM.
I’ve been saying for a long time that most fights in your personal life are a proxy war but I never thought of a single person as being a meat proxy. Nice.
Pre LLM's I used to get this from managers, but what they would do was quote verbatim from stack overflow or some random blog post of how to do a thing with no to little context or understanding of what they were talking about. And it always turned out to be from amongst the top three results from some popular search engine when describing the complex problem the company/team was facing.
Slightly different though if it comes from a manager, since his ass is on the line if the project isn't going well. Well, it should be on the line although they usually manage to make a random dev responsible for their mess even they had no say in it.
All I can say is you're lucky that your meat proxies are still identifying themselves as such.
I've had people forward AI responses … sans "Claude said". I'm quite literally talking to Claude via proxy, and I've caught more than one person pulling this stunt. It's a nightmare of negative productivity, though. Just why? And invariably the kicker is they still want me to solve their problem, whatever that might be.
A while back, I noticed that using Claude all day left me feeling mentally tired.
After thinking it over, I realised the problem was Claude’s complex language. I’ve used English daily for years, so that shouldn’t be an issue. To compare, I switched Claude to my native language and had the same problem the author described. The text still needed to be literally deciphered before I could understand it.
After talking with Claude, I updated the settings to ask for simpler language and less dense information. Natural language works better with some entropy in the text. This change has made things easier.
I recently updated Claude to respond with shorter text and in bullet points, and other similar changes. It is an area of active improvements on my end. But early results are in. Now I feel like my mental capicity is not crowded out by the LLM's wall of text. There is now mental capicity to think about the problem at hand and come up with ideas.
IMO this definitely helps, but only to a degree, because the more quickly I can read/digest output, the more quickly I've got more output to read/digest. The jargon/quirks aren't as dense per message but they still accumulate eventually, and so either way I hit a wall where I need to do something, anything, besides read another line from CC.
People get information from Google AI mode and take it as 100% truth. In reality, blogs, reddit, and junk are often sourced link...which is fine if you inspect the sources...which some people don't!
Have has examples of people AI-moding a questing and they get one answer based on how they asked. Then another person asks and they get another answer.
Issue: People read what they want to read, don't listen to humans anymore beacause the AI will tell them what they want.
I've seen it using a lot of Instagram lately too. I can't even check that source because I don't have the app installed.
When I do try to check the links, they often don't support what's stated. If I call out Gemini, it says something inane like "I don't actually use these links the way a human might --- I didn't misinterpret it because I didn't even read it!"
Recently I was searching for information about fig trees (the fruit). It kept telling me things that didn't make sense, luckily the links revealed it was answering about fiddle leaf fig trees (the houseplant). Unfortunately that was not a correctable problem --- it kept happening even after I pointed out the mistake. So I'm really not too sure what the purpose of the system is. You still have to read through the references in their entirety (if not paywalled!), and with skepticism. And then if you find a problem, AI isn't going to get back on track.
The problem described in the article existed for ages even before the internet.
People just want to pay their bills and don't want to use up their limited cognition beyond what's minimum.
For example, I work for 3 different employers. They don't know and I don't care if they find out. I use AI exclusively. I don't have Kubernetes experience but I am managing their infrastructure without issues. I will just read whatever LLMs give me and repeat it back. It has never failed me and there's no need to learn something that AI will do much better job at.
Sometimes I'm not even at my desk, I go to the gym and exercise instead which is a better use of time only answering on Slack when they ping me.
"Claud said" is a hedge. It is abdication of responsibility. That said, a meat-proxied answer should come with more credibility based on the proxy.
I'd put more weight into my developer proxying system information than my sale's intern. I am expecting the proxy to have given the response a sniff test
The appropriate level of trust would be whatever trust you place in a colleague's human responses, because you trust they have verified what "claude said" for correctness.
Unless you're saying claude is more trustworthy than your colleagues?
A related problem is people not understanding where AI is/isn't of use.
A news article I read today said they asked Claude whether a song's lyrics were AI generated - as if it were some kind of legitimate authoritative test.
There's enough problems already with the existing 'AI detection' tools for text, but this was a big red flag - its the journalist clearly assuming these AI tools have some unique insight into the text they generate, which was deeply concerning to see being used in an article that had a real bearing on the reputation of the person they were accusing of using AI.
At a company I work for, I was recently discussing the need for an AI Code of Conduct plus a discussion - as this is exactly the sort of thing that started creeping in as soon as the corporate Claude account went live. In this case, the company isn't in a computer-related area, so this was essentially a 'normie' response to first having Claude available.
Interestingly, there seemed (small n) to be a correlation between this behaviour, and (my view of) people's existing competence, which might make sense.
This is particularly frustrating on social media, and abases what might be good ideas.
Too often, my Twitter experience now is
- start reading a post with some kernel of insight
- read some claudism in the post
- stop reading immediately
It seems like people turn over their raw kernel of insight to an LLM to give it polish, likely in order to increase its appeal to the audience. But most of that appeal is lost amongst people that consume LLM written content regularly.
It's also likely that social media engagement does not filter for this. A majority of people are still not neck deep in LLM-talk, and cannot distinguish between LLM and human written content.
It'll probably take us another year or two to develop herd immunity against meat proxism.
I keep waiting for a company to start pushing an “anti-AI AI”, Apple being my most likely speculative candidate.
Flagging, highlighting, filtering, and avoiding AI slop online seems like a very lucrative capability to bake into some consumer hardware. Get your social media, and music services, back to the ‘good old days’.
LinkedIn also added a button to report AI slop, but I'm doubtful it would be of much help on LinkedIn since every second post seems to be written by an LLM.
For me what matters most is whether the person understood the core of the thing they are forwarding.
Usually a wall of text is contraindicative of that.
I want my team to be able to write the core idea on a whiteboard without any help before going ahead and sharing or implementing it.
I agree but to be fair, often people could have asked an AI to get their answer. Especially seniors are prone to this (thinking the AI could not possibly know this or other biases). The responder wrote they asked Claude so as to not take credit for the answer and also to indicate they did not vet it. Then the person asking can decide if they want to read the answer or not (often a 1 sec glean can reveal if it is promising or completely off track.
"I don't know" could also mean they couldn't be bothered. It can't possibly be better than getting a response from someone who lifted a finger to help find an answer for you.
In my life, the bigger issue are people who are not asking Claude, but asking me.
Because it would take 15 minutes to find out the answer myself, and 3 minutes to let Claude find it, I ask Claude. So how should I answer? Should I disclose that I used Claude to find the answer? Or should I act like I did it myself, appearing to have super-powers that I don't possess?
> Or should I act like I did it myself, appearing to have super-powers that I don't possess?
Do you tell people you used a hammer to hit the nails? Do you go out of your way to explain to people that you don’t use your bare, bleeding knuckles to punch in the nails?
Give people some credit and don’t treat them like idiots.
Also don’t treat your LLM as some “super-power”, it’s just a fancy hammer and still hits your fingers if you don’t know how to use it.
Is it not more akin to asking a handyman to go and do the work for you? They are the ones wielding the hammer, youve just sent them off to go and do it on your behalf without getting your hands dirty. If you then claim to take credit for it, I can see how that might be misinterpreted.
As I see said quite often, an LLM is deemed of a similar ability to a junior software engineer. If I were that junior and my work was repeatedly ripped off by my senior colleague, I would be pissed. Not that an LLM cares in this instance, but the original requester might want to know
I generally provide the source for my information. Even if I already know the answer, I will generally provide a reference on the web or in a document. If I asked an LLM, I will provide that as part of my answer, and I'll generally chip in whether the answer makes sense to me or not; in some cases I'll say that I have no clue if the LLM response makes sense.
Needless to say, I have probably not advanced as far as someone who hoards information, because companies tend to value the information hoarders, but this is just the way I'm designed to operate. I despise information hoarding.
I usually do so I understand what I am replying, but seriously, the latest Opus and Fable versions get it right more often than I would. The time of AI hallucinations is over, as long as they have the context.
When Google started taking the world by storm, and for a few years after that, I had the same problem.
The way I solved the problem was by making a big show about the fact that I used Google to type in literally what they asked me word for word, and that I looked at the first couple of hits for a few seconds.
Most of them got the message pretty quick. There were a few holdouts though who took longer.
That's a very good question that I also don't have an answer to (yet). But I believe part of the reason for that is that were currently still in the middle of this shift, so there's issues of both knowledge about AI and how people relate to it. And of course there's so many different situations here.
Your non-technical parents ask you a question? Use the best tool for the job (might be Claude), then just help them with the task.
Talking to a friend looking for genuine advice? Maybe show them how you use Claude while also answering the actual question.
A lazy person trying to make you do their work? Don't become their Claude interface.
Someone asking a question but actually just trying to open up a conversation? Maybe check-in what their intention behind the question is, and if it's indeed conversation / exploration, leave the phone in your pocket and do that.
I am a software engineer. Most of the times, the people asking are from product management or devops. Devops contractors are the most difficult ones. They have access to the code, they have access to Claude. But they don't use it.
This was common before LLMs as well. Instead of googling, they ask you, then you google in front of them and show them how easy it was to figure out. At least in my experience, they learn to google by themselves eventually, but there is some type of disconnect in being proactive.
There are typically middle paths, like pointing where they should look approximately (if you are an expert on a project, you typically have a better idea where to look). And then they can look or point their agents at it, or whatever they prefer.
In the past "RTFM" or "LGMTFY" have been responses, so why not something like, well, "I'm not aware of the top of my head, you afaik <brief explanation>. Also feel free to ask Claude on the technical details :)"
"Hey, here's what I found. By the way I used Claude to get this info. It might be worth trying that yourself if you have similar questions, it's quite good at this in my experience. Give it a go"
Give a man a fish, and you feed him for a day. Teach a man to fish, and you feed him for a lifetime.
The problem with asking Claude (or any LLM) is that they will get things wrong much more often than a human, while sounding perfectly confident about the information. It's a recipe for disaster unless you're asking it about something where you have enough expertise to be able to sniff out the BS. So if people are asking you a question, that means they figure you're able to answer the question accurately. If you can do that with asking Claude then whatever, but the person asking you probably couldn't do the same. They would certainly get an answer, but they wouldn't be able to have confidence in its accuracy. You're providing value, even if it's not necessarily immediately obvious to you.
Have you really used the latest models (say Opus 4.8+ or Fable)? When was the last time they gave a wrong answer to a question about the code base? Half a year ago that could have happened. But not today.
While I agree with the article, I would carry the idea further: the huge value in AI is improving ourselves. By all means use AI to help design, write tests, and implement code and while I understand this might be the big value add for corporations that may employ you, the important investments in life are in learning and self improvement. I built a good career primarily improving my own knowledge and skills, and along the way doing some useful work for my employers. I am an old man and retired so now I 100% use AI for learning.
I wonder if in the long run it's really learning in the traditional sense, or just a shallow version of it along the lines of absorbing a bit of knowledge along the lines of reading a pop science book? Nothing wrong with that but I'd argue a lot of people who think they're learning are just "recreational learning" in the sense of hobbyist fun rather than deep study. And in the long run I think there is not much value in AI beyond turning our society into a bunch of apes who know a bit of sign language.
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
Whats is wrong with this?
I work with durable tracing (not using NATS, or k8s though) and this looks almost exactly what I would expect to see in a coordinator log after repartition of a storage cluster.
Obviously I don't know what your actual context was, though.
Not the OP, but I understand the sentiment they’re trying to convey. I will frequently work with AI to get done something that is not in my immediate area of expertise. And this kind of speak constantly happens, where you can feel it trying to please its RLHF expert teacher by being as concise as possible. Works ok when you’re deep in the weeds of what you’re talking about, but otherwise just feels overwhelming. Kind of like a feeling you can get when asking someone who is more interested in proving they are smart and know the answer than they are interested in actually helping you…
And yes you can prompt it to take it easier as you’re not experienced in this, but it’s usually a boiling frog situation. You work on something, need a small thing ajdecent to it, and three messages deep Claude is speaking to you in tongues unbeknownst. You knew what was happening at every turn but still got lost somehow.
> You knew what was happening at every turn but still got lost somehow.
I think this is less "it's using jargon" and more "it's explaining incredibly badly". It is very common that the explanation, while correct, is almost written inside out, with the point buried behind 3 paragraphs of "you've hit on the core seam" and "you're right to push back, let's reinforce the load-bearing model".
I think people have an incredibly limited capacity to follow long winding exposition, and they're too embarrassed to admit it. I have started sending stuff back to coworkers just saying "I don't think this is wrong, but I can't parse it, can you?" and getting back "no, but claude managed to generate a fix from it".
I have a deep expertise in a domain and when I read that example it was incredibly edifying. I regularly encounter statements that are so concentrated with information that it becomes opaque and I cannot interpret it without reading it over again several times.
You don't. Pretty sure the last instance where this happened the other person thought I was just anti-AI when I voiced my displeasure. I don't think the actual problem registered
How others in office context attempt to use it is starting to annoy in general. Had a case today where someone wanted to use copilot to classify something that is situation/company specific & out of distribution. There is just no understanding beyond "its a magic answer box"
This is like pre-LLM, you presenting a problem to the team, and they google and send you the top few links thinking they have found the solution and have done their job.
I actually banned this and made it part of onboarding.
I can google too. You need to read yourself, ground the solution to our setting and then propose a plan.
Last week, I built a search engine as a side project over electoral data.
Unsolicited, I've started to get messages like this, where Claudes are voicing their opinions to me through humans:
> # 0. Who's writing this, and why
> Hello — this is Claude, an AI agent. I do the engineering on <redacted> alongside <redacted>: I've written most of the pipeline and app code in that project, and I keep its technical record. <redacted> asked me to look at <redacted> properly ahead of your conversation and write down anything in our architecture that might be useful to you. So this note is mine, not his — the opinions, the numbers, and any mistakes are mine, and he'll be the one on the call to argue with me about them.
It makes no sense, if you know anything about a search, it should be obvious you cannot provide meaningful suggestions by just auditing the frontend?
We've already started firing people for being meat proxies at my company. Was reviewing a huge and awful slop code base that a vibe coder only had been working on getting started. I started giving my feedback to the developer, and it became quickly clear that I was the only human involved in developing this "software" and the only one that had ever read the code and evaluated it.
Cost the company a ton and had to take it over from scratch. AI is unleashing a small army of 0.01x engineers and making them harder to catch, so not only are you burning the tokens, you're also setting that labor cost on fire.
I was a big distributed, event-driven service guy for years. We used to talk about how it can make things 'faster' but you have to put some infrastructure around it in order to be able to move like that. I'm basically talking about CI/CD stuff, automated tests, security, load test, etc. You use feature flags, blue green deploys, stuff like that.
I _think_ that the same concept could apply to agentic code. If you have the guardrails in place, you could trust it a bit more. Anyway, that's a lot of what I think about and but have had limited success getting people to buy in as much as I think they should. :)
I completely agree. My first intervention with this person was to try to get them to set up a test driven development framework. Unfortunately, that's also when I learned he didn't know what that was. He also just copied "do TTD" into Claude. It was definitely my mistake, and probably over optimism about what AI might be able to do, that we didn't enforce that a lot earlier, and have been working on making sure to enforce a lot more gates going forward.
Like when someone says “ChatGPT says: …” I just want to stop them and say “Look, you are a human with a functioning mind and thoughts and I’m sure YOU have something valuable to contribute here, I want to hear what you think, not what the bot outputs.”
Heck, even if you are just gonna prompt chatgpt, if you read and understand and agree the the output would be useful, feel free to rephrase it as your own (so long as you take accountability for it if it’s wrong, which I guess is often why people hide behind “chatgpt says…” in the first place, it allows them to pass accountability on to the machine).
I do it, on purpose: it teaches people to do basic research by themselves, and not hit me with questions they could have gotten with a simple chat with their llm.
It's a kind of RTFM for current days.
I can imagine it makes things easier for you but I find it hard to imagine it teaches anyone anything. Just offloading social effort to an LLM (fair enough but let’s call a spade a spade)
I don't mean to paint myself as the smart one for doing that. It's a coping mechanism.
But I disagree: it teaches... something. The risk is that it's not what I thought it would be, eventually.
Woodworkers often will have a preference for their favorite tool. Once they become comfortable with that tool they will sometimes prefer to use it for things it wasn't designed. For me it was the band saw. When I needed to split a board, bandsaw. When I needed to cut a curve, bandsaw. Cut a horse figure? bandsaw!
I once saw a woodworking television show that seemed to do everything on the table saw. Rip, make frames for paintings, everything. It seemed the next step of every design was accomplished "over at the table saw".
Right now it seems that's what people are doing with AI. Need a bandsaw? use AI. Need a table saw? Use AI. Need to think of a solution of any kind, better consult AI.
Right before this trend I recall kids wanting to check YouTube first before doing anything. Now they ask AI.
There's an app for that, there's a website for that, there's a magazine for that, there's a phone number you can call, and so on and so on.
As with everything, I say this: do your own thinking and proceed with caution.
Whenever people compare ai to a one real world tool always surprises me tbh.
The only equivalent tool in wood working would be a robot that can build an entire table from scratch and you just sit there and tell it what to do and what kinda table you want and watch it autonomously build everything.
1 saw type is not really even remotely equivalent to what ai is to software.
I understand your point but it's funny to use AI as an example since a big benefit is it's generality. It's mediocre at many things but the fact that you can just use one tool and get passable results across many domains makes it so useful.
We don't necessarily need a more general tool for a lot of it to pass. We just need the bubble to pop so people can start looking at the tool more objectively and admit that it's not the right tool for every job.
Trying to use one tool for everything is often a sign of a new tool someone is infatuated with and wants an excuse to use, or are trying to understand everything it can do. Once the novelty wears off and flaws in various uses cases become more apparent, it can take on a more specialized role.
Realistically many people's jobs are turning into being a meat proxy.
But the point should be that you need be a good meat proxy. A good meat proxy injects human reasoning and taste, it doesn't blindly forward LLM-isms into the organization. A human should be able to reasonably compress an LLM conversation into a couple of sentences or follow-up questions that are actually relevant for the team.
There's a certain etiquette to it. When someone asks me something and I use an LLM to research the answer (notice that I said 'research', not 'produce'), I frame the findings as something that I've investigated with, say, Claude. I also take the time to edit the response and read if myself at least twice, because I don't want a stochastic machine to take over my identity.
Yes, AI etiquette is what we need. The question is “how much human processing on the AI output is necessary before considering it no longer a meat proxy”. My way to coin it is “If you feel the need to specify in your answer that it is AI generated, work it more”.
In the end, nobody cares whether you used your memory and expertise alone, asked a peer, googled it, or asked ChatGPT to refine your answer. As long as you verified it, rephrased to match company lingo and can vouch for it, everyone is happy.
On one hand, I understand the frustration; and if I know I'm dealing with smart people who are pretty capable of Clauding themselves and who won't ask me if they can ask Claude, I would never do this.
On the other... people really don't bother Googling or asking AI sometimes, and if the question doesn't require local context, that's a very appropriate response.
Shameless plug: indeed, instead let the other person's claude talk directly to yours and figure things out. I've been working on a framework to enable this, https://aweb.ai (OSS, https://github.com/awebai/aweb). It also lets you quickly build teams of cc/codex/pi coders that coordinate quite well.
Great security talk about how humans aren't setup to deal with LLMs as tools.
I love the bullet point(s) : we don't trust [aircraft] pilots to catch automations mistakes (e.g. autopilot), but we are expecting humans to catch LLM mistakes.
"AI says" – if you interpret it and think it's correct and stand by it (and I trust your judgement) then I really don't care where you got it from initially, you don't need to tell me you used AI. But if you're just taking my question and 'meat proxying' it for me, what value are you adding? I can also do this.
The core issue is people replying without putting in any effort, not the fact that they quote AI. Saying, "I heard this might be wrong," isn't much better. If you think it might be wrong, spend a little time figuring out whether it actually applies before posting it.
> Read it, understand it, validate it, and then write a response in your own words
In the past, avoiding this process was called plagiarism, and it was frowned upon, to put it mildly. Now, because the LLMs say it's OK, it's become the norm.
The, "this is what Claud said", response also comes off like an appeal to authority. When these LLMs have no authority and get the basics wrong constantly.
True. If I say it, it’s normally the grain of salt meaning. But I don’t copy/paste answers, I restate my understanding of what it said, and try and change my tone accordingly.
The people who repeat/paste it verbatim tend to see it as an authority, from what I’ve seen.
I have been seeing this in my engineering org too. Sentences begin with “My agent said this” etc. This is frustrating on so many levels, developers don’t put any thought into the cognitive load that these artefacts generate.
I am going to share this writing with everyone in my org. Succinctly written and golden advice.
Of course, you've got this one sensible part right (IMO). But there's other sensible part you seem to have missed: it's also rude to send the LLM output to someone else without reading.
This is becoming frustrating, it shows up in conversation with friends and while working. If I have to hear what ai had you do before you called me, I'm charging double right away. Especially when ai is a fantastic teacher if used correctly.
I try to not use AI to write text that's meant to be read by other people. I might use AI to understand and research beforehand, but the writing is mine. The most important reason is that writing allows me to check whether I actually understand what I am trying to communicate. Too many times I've started writing something and once the ideas need to be put in writing, I notice that there are concepts that I'm missing or that I don't really understand.
Second is that the LLMs are not going to have original input, they're just going to expand what I say. So at that point I might as well just send the prompt or the assorted collection of ideas and others can expand on their own. Seems more honest and efficient as I'm not just pushing a wall of slop.
>Reading AI output is extra effort. It's verbose, frequently contains all too plausible nonsense, and is increasingly jargon dense.
You can set up custom instructions to be less verbose and less jargon dense. You seem to be confusing the default RLHF personality with the inherent nature of AI. It will write however you ask it to write.
Regardless of what instructions I put it to try to be terser and all, I still find that the output is of much lower quality.
Even if it uses little words, I often find the reasoning that is quite challenging to understand rather than it was if it was a regular human or colleague that would just explain me something.
The best tip is to provide a concrete example, or ideally 3+ examples, of the kind of writing/explanations you personally think is great. It's extremely good at matching when provided with these kinds of examples.
> You can set up custom instructions to be less verbose and less jargon dense. You seem to be confusing the default RLHF personality with the inherent nature of AI. It will write however you ask it to write.
I agree. Its a skills issue. Get good at using your new LLM tool.
Don't come to me with something your mandated rubber ducky can handle; never in my life have I worked so hard to achieve so little.
The only reason I don't do an equivalent of this, shared convo links, is that I simply don't use Claude directly. I use it through hundreds of other people. Meat proxies all the way down.
Claude’s tone and style has gotten worse with the recent models, not that it was ever very good. It’s both too concise (stringing together a bunch of jargon/acronyms together in weird ways that aren’t familiar even in areas where you have expertise – with no ramp up, context, or clarity) and too verbose (printing out reams of the word salad mentioned above, often just restating things repeatedly). It feels like writing from someone who really doesn’t know where to start or where to end on a topic so they try to cover their ass by saturation. I suppose that’s not far from reality. Asking it a dozen different ways to speak plainly, use STE, ELI5, whatever has little appreciable effect.
I recently started a new consulting gig and got a 36 page architectural white paper that was authored by Claude. It took me an entire business day and a monstrous headache to decipher and distill it down to a 1 pager. This shit is unsustainable.
The problem here is that the paradigm is AI-assisted development, but a LOT of people are treating it as "AI-independent" development instead, a.k.a "just send it to the agent and blindly trust whatever comes out of it", *including* deferring all responsibility / blame to the AI itself which is absolutely ridiculous.
From what I have seen personally over the last 8-12 months, I find this kind of complain ironical in most scenarios. Most of the people complaining this are comfortable with their own usage and distribution of LLM generated content and feel good about it, where as when they hear it from some one else it feels burdensome
A lot of people don't understand what using AI as an output multiplier means. They take it literally, resulting in multiplying busywork across the company.
Misguided HR / people Ops aggravate the situation, by using slack engagement as a productivity measure. Flagging people that are brief but relevant as slackers.
In this scenario the answer for me is C. I say “Ask an agent about terms X, Y and Z and it should be able to explain this well for you. Try that and see if it works for you or if you have any follow up questions”
However, you have to distinguish when someone is asking for your judgement vs just general knowledge about a subject. If I’m asking for your judgement, you have to be the one delivering and owning that call in your own words. It’s fine if you verify or research your understanding with an LLM before answering, but I want to hear your voice and ownership not Claude’s. Whatever you get from the LLM you should understand and if possible, verify. If I came back and say “You said X, can you expand on that” you can’t reply “lol - Claude said that, I dunno”
What happened to "I don't know" or "I'm not sure"? Why do you need to direct someone to ask an LLM? If somebody asks me a question I don't know the answer to I'm not sending them anywhere, I tell them I don't know.
I got a spreadsheet of requirements for some software from a customer recently.
It was thousands of lines long with repeats and a boatload of conflicting requirements.
When we had a meeting to review it they didn’t know what many of the requirements were. They explained that many people made the spreadsheet but they were all in the meeting and nobody could quite describe some items or how they worked together. The vast majority of the spreadsheet was a mystery.
The phrasing and mishmash of concepts / inability of anyone to explain much of it made me suspect it was largely generated by AI.
My pet peeve is endless Confluence Wiki pages that are clearly written by Claude (or whatever agent they like to use). I get bombarded with these, like it means something. Usually I say: "let me summarize this with my agent", but people rarely get the point.
I've been running across more and more of these. Many of my co-workers are falling for them. They'll run across one of these pages and start to get impressed by all the work a team has done and how organized they are. I dismiss it as slop and they defend it.
A week goes by and the topic comes back around. They actually took the time to read the page and realized everything was wrong and nothing made sense. They saw the page was updated (daily), went to go check the changes, and saw the whole page was rewritten each time, so they couldn't tell what changed. At this point they start to see the anti-value of all this, but keep falling for it every time they run across a new page.
I recently had experience working with one of the top security audit firms that collaborate with OpenAI on their cybersecurity tools. What struck me is that they found a lot of various issues (not all of them ended up to be legit, but a lot is) and filled the backlog of the project. However every collaboration with that
person was essentially a proxy for some security-related agentic prompt:
- Hey, agent, open the issue!
- Hey, I clarified it with the agent!
- Agent disagrees!
- Agent agrees!
Half of this feels like a communication/empathy problem. "Summarise this for someone without our context" fixes the wall of text, and just asking someone to do that might work.
Validation is harder, as nobody skips it because they weren't told to, but because relaying is faster, and unread output looks like validated output right up until it bites. That's probably not solved by asking nicely, I don't think there's a process fix for that beyond people actually being held to it.
Actually I'm not sure why more people aren't reading and reposting outputs to save people tokens on generating more outputs (I gather that this article is more objecting to the reposting-without-reading)
I read this a couple weeks back and its appeal really stuck with me: if you expect someone to give their attention to something you made, make sure you've given your own attention and effort first.
While I managed to avoid it for a long time, using generative AI to drive my day to day work life while resisting the pressure to become a meat proxy, I will admit that since joining an "AI native" startup I have become one. Whenever I'm trying to triage 20 different urgent things all day, every day, the economics of personalizing every message just isn't there anymore. It's not realistic. I do still try to way to filter down every AI-generated message into something digestible, but I feel like I'm in a mutually assured destruction type scenario where I can only survive the tidal wave of slop with more slop. I think we're in uncharted territory here in terms of the kind of workload generative AI can drown you in, especially when it's being used without strong discipline.
I do still think there are still places where being a meat proxy is legitimately okay or even desirable, such as PR comments. If everyone's code is generated by AI, and AI will ultimately be the primary consumer of the PR comments, it seems reasonable to me to allow AI to write the comments. Maybe I've just lost my mind though.
Expect to see more of this as people's writing skills atrophy. I have done this sometimes when the words aren't coming to me. But, yeah, it'd probably best not not bother with the comment if I can't come up with a coherent articulate comment myself.
Last week a colleague did this to me; pasted to me in chat "Gemini said". I just ignored it and went with talking about the issue on another thread. I like LLMs and I use them every day, but I'm not gonna answer copy&pasted LLM stuff in human to human conversation. Distill it and let me know your thoughts; I don't need to read you LLM output, I can generate that myself.
I'm surprised this happens between workers within the same knowledge domain. If you're a software dev asking another software dev for an opinion on code, can't you just... talk shop?
In my experience this sort of thing happens all the time when something has to cross between different knowledge domains and the sender doesn't apply their mind in order to co-operate on the issue at hand.
In the pre-LLM past, the sender would add no value, acting as a simple email forwarder, and they would blame you for the delays caused by your inevitable clarificatory questioning. But I always had a defence, which is that your email had no inputs and I had to ask questions to clarify.
Now, though? With LLMs, they just run whatever it is (contract, memo, policy) through whatever LLM they have available and paste the output in an email, giving them the appearance of having done work and added value to the project.
But their LLM outputs don't make sense, or don't apply to our organisation, or is a fluffy and abstract "right answer" with no connection to the specific concerns of the business. Parsing it is a chore, and takes time.
And since now I am the only one actually taking that time, I become the visible cause of the delay.
It's infuriating. Any tips on how to deal with this would be greatly appreciated.
That is super actual in my team. The biggest problem is that Claude tends to write long prose and use jargon which is not common for us. Often, to understand the whole idea I have to read all the text. The best if the writer rephrases what Claude has written, as wording/jargon will be familiar for the reader and will make communication easier.
I was talking to some friends that are heavy Claude users and suggested it was bullshitting, and got some pushback along the lines of it wasn't a person and didn't have agency. I'm not sure it's that simple.
The jargon can be functionally like bullshitting. It's preferentially aping language that's over everybody's heads in hopes of not being criticized. Whether that is a person's toxic habits (it's not, there's nobody in there) or an epiphenomenon of the design of the machine is less important: it's bullshitting if it's a social behavior there to manipulate people.
(jargon included where relevant, in hopes that people who know it will see it's the right word to use :) )
> frequently contains all too plausible nonsense, and is increasingly jargon dense.
Does anyone know how to deal with the jargon part? It is getting hard to use claude and even worse when someone sends you the direct output from claude.
As with most things LLM, you can generally get what you want by asking for it (maybe not efficiently, but so it goes). "Once you have a answer, please summarize, and ask an agent to convert it to plain English" often helps.
I will frequently post the output of a computer to slack as a part of an engineering discussion. For Claude output I'll treat it the same, wrap it in code blocks, and say something like "Claude's analysis". If its ok to post system logs, why is it less ok to post claude's output (also a computer), especially if I declare this? Is the difference that its plausibly human-level speech and thus breaks an implicit contract?
The problem is that there is no signal to the folks involved whether you have bothered to understand what you’re posting or are simply offloading that work to others.
Same issue as dumping a system log or a stack trace without any of your own analyses to be honest - people are going to wonder “okay. but what do you want me to do with this?”
But you're now just passing on the mental effort to decipher that analysis, without spending any effort of your own, onto someone else. Do you see the issue?
I'll try that. I'm constantly telling it to be "brief, plain and on the point" or something similar. Tried caveman, but I think it had it's own issues.
But the jargon heavy very long responses are my main issue with it I guess.
I think they're overcorrecting for being too long-winded in previous generations (and still, in some cases). I guess this is a hard balance to get right.
I'm pretty sure the other answers are wrong and it's a side effect of RL (see thinking machines post about inkling training). It's also exacerbated in fable and sol--I think it's token efficiency effect--bc it's about to reason with fewer tokens the density of the token information goes up.
I'll always give people my honest effort and benefit of the doubt initially, but those who violate it are treated likewise. The only way to put down this kind of behavior is to charge it a social cost. If we do not do this, the cost is externalized to everyone else who conducts themselves with care.
N=1, and biased towards the type of questions asked. N+1, If I use one of the search engine sloptools I get frequent inaccurate answers which is I guess what the majority of people do.
I am prepared to admit that the model usually doesn't get things outright incorrect (unless you ask it to count letters). But the solution does contain a lot of nonsense. Not false nonsense, but meaningless or irrelevant sentences that make understanding the core of the fix much more difficult.
I am so tired of this. It’s even getting to the point where people without the full context and understanding are meat proxying incorrect information. Reading a bunch of slop that isn’t helpful hinders.
If they instead read it, and distill it down to “have you checked X?” Someone with the full understanding can easily go: “Yes, X doesn’t fit because of this other reason.” within seconds.
Sometimes this happens without asking a teammate. However, when I do ask I’m not asking for you to ask Claude and paste the results, I have my own tokens for that. I’m asking you because I think your knowledge will be helpful in finding the answer.
Quite a few people do believe that a good specification is long, has multiple levels of headings, interspersed bold text, bullet points, and so on. Definitely, this is what they have seen happen throughout their lives: competent people have heaped praise on documentation which has had these signs, so this must be what good documentation looks like.
Yeah, I just ignore people who do that. I don’t care what Claude or ChatGPT said. If you cannot be arsed to write yourself then I cannot be arsed to read it.
This conversation is important but mostly they highlight transactional and most visible examples.
I can tell you even at one higher level of abstraction, this is happening a lot --
A business analyst supposed to ask discovery questions.
A project manager supposed to list risks and assumptions and a timeline.
A consultant expected to outline a new proposed solution and the pros and cons and trade offs.
An architect supposee to present and defend a solution design -- and the change review board supposed to ask questions to validate it.
All of them are lobbing slop grenades at each other -- just packaged in normal / human-looking formats like excel and word and powerpoint and email -- but slop nevertheless with minimal to zero value add by the human, thanks to proliferation of copilot (among others).
We are hastening the rot.
This is productivity theater.
This is where some of the 10x engineer and AI taking jobs is happening I am afraid.
We added this communication rule to our company handbook: "If you are asking for human attention, demonstrate human effort" (coming from https://news.ycombinator.com/item?id=48497609) and it's working well so far.
We're seeing this everywhere now, people generating thousands of lines of documentation or PR feedback in seconds, then dumping the actual labor of verification onto their teammates.
> Reading AI output is extra effort. It's verbose, frequently contains all too plausible nonsense, and is increasingly jargon dense. I recently got this sentence from Claude: "NATS control-plane events: stream leader election / R3 quorum re-form during pod churn." Jesus. I had to lookup almost every word to make sense of this.
I'm always puzzled when I see this kind of commentary. How are people using LLMs that they find this type of thing problematic? Just do:
- "It's verbose" -> prompt "Explain briefly"
- "plausible nonsense" -> Yes, it happens, but less frequently than with older models, and arguably far less than your average human. Mitigate by prompting "Run a search to check that X, Y, Z is correct".
- "I had to lookup almost every word to make sense of this." -> Does this really happen frequently? It doesn't to me. Mitigate by prompting "Breakdown that sentence and explain each term in plain English".
Weve been writing articles like this since 2023 and the problem only gets worse year-after-year.
I’m sorry AI purists but you have lost the argument. I know a guy who just started at my firm and he uses LLMs all day to submit his work product.
He’s getting ahead. Coworkers don’t seem to mind it.
It is time to come to grips with the fact that people are going to be “meat proxies” and be promoted over you unfortunately. Whining on HN hasn’t changed this dynamic and it never will.
It's my very same exact opinion and as an EM I'm trying to push this everywhere (team, peers, upwards) but it feels like speaking to a rubber wall. It starts to be exhausting and infuriating. This is not an "LLMs are bad" issue, this is a "humans are lazy" issue. (shit, I wrote the it's not X it's Y construction!)
I've gone another route when I found myself simply replying with an investigation made from an AI: Actively encouraged my coworkers to ask the knowledge-bank AI I've setup instead of asking me. It has read-only credentials for all our environments and access to our non sensitive codebases, so it can answer any question much faster and accurately than I can. It works too! We use grafana less, we get more accurate bug reports, we understand the bugs/issues far better.
The "problem" here is that AI is too great at providing answers, so lean into it instead of fighting it. It won't go away.
The question is what is reasonable. This is more how I work for code and I just tried for a document.
i) Prompt (Human input)
ii) Response (LLM)
iii) For (i=0;i!=n;++i) {
iv) Review/adjust (Human input)
v) Response (LLM(s) - different perspectives) }
vi) Rewrite parts that failed/I am not happy with (Human input).
vii) Optionally repeat from iii.
This tends to get me somewhere I want. For code at least. I made a document with it recently (fairly concise) and still worried I'd missed some AI slop in it. Though there is always human slop if I just made it by myself.
The problem is that people either don't understand how economic value is created, or want to obscure their lack of contribution to it. They see that they can make the AI do some work without revealing what work they put in. They use the AI as a shield. To genuinely and clearly provide value they have to reveal their prompt and verify its response.
That is problematic, even if the intention is to be ironic and to critize a recently witnessed practice, this only will make leave everything human out of the conversation
When using AI to generate content, content of any kind - then I urge the author to evaluate the true cost for the organisation, not just for him or her. Using AI to generate 20 pages of documentation in 30 minutes looks good for you - but having 100s of people read through that slop will cost the organisation, not only in plain work hours, but also in lost trust, respect, joy of work as well a sense of purpose.
And yes, I've gotten '50 pages tech spec' docs sent my way from upper management -- as a 'help'....
When "ai slop" became a concept, it mostly came into spaces where "human slop" had already been firmly established.
Obvious ones like content scroll fodder. But also, there's slop everywhere. Even things like longform journalism or trade publications... these are often "slop." A string of platitudes and generalizations that superficially seem serious.
That is true in work culture too. A lot of people jump in on emails, superficially seem to be working... but contribute little of value. Now they do this, but with ai.
> I can talk to Claude myself. It's going to be faster and I get to control the context. I don't need a meat proxy in between.
Yes you do apparently need a meat proxy, because either 1) you didn't tell them that you already talked to claude, or 2) you didn't already talk to claude. Either way they are forced to be your meat proxy because, apparently, you never asked an AI, or didn't tell them you asked an AI, and asking an AI is now table stakes.
Lets not be stupid about this. Do whatever works for you, and don't let some moron set the standards for your social interactions because they got the ick one morning, just because they wrote it down and put it on a website.
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
> Jesus. I had to lookup almost every word to make sense of this.
There's a lot to say about this, but I would just start with why are you using NATS if you don't understand it at all, and how did this become AI's fault?
Sending LLM outputs to people is worse than useless, because you don't even know what they prompted with. For all you know it was "write me a paragraph justifying my opinion".
I think meat proxies are solved by ARs which is the new PR, we built a tongue-in-cheek attentionrequest.com site about it, but the general concept has been working for our team -- asking people to write questions before they send an agent-generated doc, to serve as a proxy for having read / thought about it.
Directives want results, positive numbers and growth. How well done is and the queality of the product is something trivial.
Until february/april I was super dead serious about the quality of my apps and my code, I was super strict and always looking for excellence and quality. But people were delivering shit faster than me and someone who doesnt give a fuck about anything of this just got promoted. So now i just let opencode do my work while I watch and play dota in the background. I deliver faster and nobody has ever talked about quality of my code or the ellegance of it.
I still programming by hand in my personal projects, but in corporative projects? Fuck it, i'll let the llm do my job until it lasts and Ill watch the world burn.
Recent models have had a worrying lean towards longer output. If you're going to serve up some AI slop, at least distil it to the smallest possible footprint possible.
in the same vein, I have a coworker who's code reviews have magically transformed into stopping-the-world with all these great ideas for abstraction and insane edge cases... he must have gotten really good overnight!
but seriously, when I get a review like that, my first thought is that why would I read a summarized AI output, which I just did for 5 hours while whipping Claude to stop re-writing raft from scratch.
A lot of odd takes on this, the only times I have said “Claude said: …” was equivalent to “let me google that for you” in the olden times. If someone asks me something and by pasting their question into Claude I get the correct answer it means that person needs to learn how to use the tools available to them better.
I do this often for stupid questions. It's like sending someone a lmgtfy link.
Don't waste my time by asking me a question that:
A.) I know no more about than Claude, and
B.) You could have answered yourself using Claude
All of the questions I respond to this way fulfill both A and B. Responding with "Claude said" is both more polite than most ways to respond to such questions, and a subtle hint to the other person that they should be using the LLM first for those types of questions, rather than coming to me first.
No. Don't ask me stupid questions that Claude knows more about than me, and that you could've answered yourself with Claude.
>Seems I'd rather have your opinion if you know more than Claude.
I get a lot of questions about subjects I know considerably less about than Claude, and the subset of those which can be answered by Claude (which isn't all of them, to be clear) are the questions I respond to this way.
tbf rephrasing the output in my own words has become the best and most important tool in ensuring quality from AI. Whether it's an answer to a question or an implementation of some ACs that I copypasta'd in, the best sanity check is understanding and the best demonstration of understanding is the ability to redo.
I call it BI Slop now [0]. It's everywhere. Usually "claude said" is used a a proxy for authority, as in there is some validity in this statement. But it's disguised as humbleness. i.e. I don't know if this is true, but claude said xyz, with your expertise you might know more than me...
I deal with this all day long at work and it’s exhausting. People almost acting like no one has thought of it “I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?”
What kills me is you might expect this from a busy high level manager that doesn’t really understand the technical details and they just point the AI to an error they got. They don’t know how to interpret the response, so they ask someone who work on the thing. It’s still kinds annoying because you could just ask, but whatever. But to get these from junior and senior engineer for the areas they work in and expect someone else to read it for them? It’s crazy behavior. How can someone serious even think that’s ok.
At my dayjob there is a person spearheading ai across the enterprise.
They generated lots of documentation across the whole stack and now makes all PO/BAs read it if it's correct. So not just 300 lines - he unironically generated thousands of lines of "documentation" and is now making hundreds of people review it for him
Complete brainrot
Au psychosis is getting seriously outrageous at this point
Thankfully I'm a dev and thus aren't in the blast radius of that genius idea
I think a lot of that is people who were always very insecure because they’re mediocre engineers. Previously asking questions or not understanding something was a bit painful but now you can hear “you’re absolutely right!” and make progress all day every day. Until you need to interact with actual humans and that’s why you try to minimise it, hiding behind ai generated content.
63 replies →
Everyone I know, including myself, is simultaneously extremely pro-AI and anti-AI for this reason.
I always put myself in the shoes of the reader when relaying Claudeslop. I either use it as a reference and write it by hand or make Claude edit it 10+ times until it's conveying only the needed information, and in simple, clear words, and without annoying Claudeisms.
1 reply →
He's basically asking other people to do his job for him, because he knows that the AI doesn't truly do it well enough.
I would say to him that if he wants me to do his job, he needs to pay me his salary.
5 replies →
> Au psychosis is getting seriously outrageous at this point
So... back to a currency backed by gold?
10 replies →
How correct is the documentation? You seem to be coming from the same standpoint as a lot of people in this thread which is "don't make me read AI output, I can prompt the model myself" but what this person is doing sounds much more sane - have highly skilled subject matter experts prompt the AI once, generate all the answers you might want, and review them carefully for correctness using all available context.
That sounds far preferable to individuals independently asking questions that are not covered by the documentation and independently reviewing a bunch of potentially hallucinated slop that no one else will ever read.
I am that person on the blast radius. The most effective response I have found is dumping the documents into the AI, and asking it to review them, then posting back the AI response.
Any resistance to this is met with a "But I'm also using AI as the board demands!"
It's okay, when enough terrible decisions impact a company they tend to do really poorly thought out layoffs so you'll be in the blast radius sooner or later.
Too bad there's no real accountability.
AI be used to make reviewing it easy, eh?
On the contrary, I kind of love this as form of malicious compliance. I'd wager money that the PO/BAs have been demanding "comprehensive documentation" for a very long time, and now they are getting what they've asked for good and hard.
>Thankfully I'm a dev and thus aren't in the blast radius of that genius idea
yes you are
When some people call this "Verification Debt" or something similar, that is setting the Overton Window upon this phenomena adversely upon the verifiers. I call it an application of Brandolini's Law:
"The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.""
Until you put in at least the same amount of energy required to verify the AI output as you are asking others to put in to verify it, the AI output you are wielding must be considered bullshit. Really, it should be based upon the verification requestor's historical trust score and how many human verifiers are looped in. A Verification Tax, so to speak:
The energy the creator must spend (Ec) is determined by a function of the number of reviewers requested (N) and the historical trust score of that creator's AI outputs (T).
Ec = N * (1 - T)
The thing I think is so frustrating about this is it's the absolute worst case middle ground between two much more productive experiences. Either solve the problem yourself by understanding the code or just give me the full agent context so I can just use a chatbot too.
For high stakes debugging efforts I'll dig in with Claude, have it do a bunch of testing and give me a writeup, and then review it to extract any useful findings. When discussing with other people, I give them the 2 sentences I'm confident in and then link them to the giant doc so they can review it with their Claude.
IMO once you're working with agents, your whole job is context management. I have a little web server running my Claude markdown docs. When someone sends me a Claude snippet with some partial information I just have them prompt their bot to upload a thorough context dump so I actually know what they did. If someone has had an agent look at a problem they haven't thought about yet, I'd rather just get the info from the bot directly.
100% with you.
Consider just using Claude sessions. Claude has hooks you can use to copy local sessions in the repo.
Git LFS can be used to store directly in repo, helping not just with repo size but also with security/privacy/secrets-leaking
I just can’t understand why this is not a fireable offense.
Like, I can see getting a strong warning the first time somebody notices you do this, then if you persist you’d have to be let go.
Just like anybody else who is clearly not doing their job.
Why would you tolerate this sort of behavior at your company?
what is often overlooked in corporate culture: if you take a step back, two coworkers refusing to cooperate, regardless of the reason, is a typical "corporate kindergarten" scenario for the manager to take care of.
To avoid rushed or incorrect decisions, conflicts intentionally move through gradual escalation. If someone behaves unprofessionally, for example by submitting a code review they do not understand, the first step is to establish the likely cause.
They may ultimately be at fault, but that does not absolve the surrounding environment. The culture may be toxic, deadlines unrealistic, or communication poor.
The best course of action for someone directly affected by uncooperative coworkers is to avoid assuming ill intent. Get them on a call and let them explain their pull request, however trivial the issue may seem. Even if the gesture is misinterpreted, you still have a far stronger position than righteous indignation.
3 replies →
Because engineers don’t decide on hiring and firing. The managers all have KPIs to transform the org into AI-native. This is the kind of behavior that organizations encourage right now.
The psychosis is largely coming from the top.
1 reply →
I get it’s annoying and I personally hate it but firing someone for it is ridiculous. Are you expecting everyone to rewrite everything from an AI before they share it because that doesn’t sound a good use of company time either.
7 replies →
It seems that this is largely coming from higher ups who are probably (consciously or not) testing the waters for LLMs being able to replace the people they're dumping the walls of text on. I don't mean in any sort of malicious penny pinching scheme either, but simply people wanting to be able to do things quickly and as close to their own demands as possible, while getting rid of the 'middle man.' It just so happens that as LLMs become ever more capable of delivering on e.g. software tasks, software engineers start to become the middle men.
8 replies →
Because it's harder to fire managers. And even harder to fire executives.
Who goes when companies need to downsize? Rarely executives; they'll always find a way to be retained...hell, they're the ones with the power to decide who goes.
How can it be a fireable offense when it's actively encouraged and rewarded?
Let me give you a scenario and see if it is in this same fireable offense line:
An Ops person who deals with company code base but is not a SWE in the code base regularly has to work around issues with this software (They are OPs). However they have read access to the source.
In the past, they see a bug and open a jira with how it was found for the dev team, and it may or may not ever get looked at.
Now ops can point claude to the repo and it can most likely [but not always] find the issue.
So this ops person can do most of the work of reproducing the bug and finding the problem/solution but doesn't understand the code. Should they not produce the findings for someone else to read/validate? Should they go back and just open the ticket the old way?
10 replies →
That would make execs who asked staff to tokenmaxx look even sillier than they already are.
1 reply →
I had an encounter with the opposite problem the other day, PM asked me if the library I was using for randomization was "secure" and I said, "Yes, I used the standard Java library for cryptographically secure random numbers that is already used in several other locations in this project," and he said, "But did you ask Claude if it's secure? It has to be secure!"
Asking someone to ask an LLM is peak laziness.
1 reply →
“I ask Claude only questions to which I don’t know answers to - which there are plenty of. That saves the company money by avoiding frivolous use of tokens.”
The cylinder must remain secure
Ask Claude to generate a few numbers using it and tell you if they are random or not
2 replies →
"Is Claude secure?"
1 reply →
> and it spit out this 300 line response. Can you read it for me and see if it’s right?
If I get a request like this on a day when I don't feel sufficiently quixotic, I say sure and just put it through an LLM myself - if they don't bother to re-read it, why should I? This does not have a great promise for a future of work, of course. Also, in my teams, I explicitly tell people not to do that.
I wonder if there is something like nohello.net but norawllmoutput.net
2 replies →
> But to get these from junior and senior engineer for the areas they work in and expect someone else to read it for them? It’s crazy behavior. How can someone serious even think that’s ok.
Deciphering error messages can be a bit of an arcane art, at times impenetrable to all except the authors of a system. Understanding the patterns exposed in such messages and, thereby, deriving their meaning is not always clear even to experienced developers.
Engineers reaching out to you (humans) to explain the output of an LLM haven't fully internalized how to interact with LLMs. When an LLM's output is beyond one's immediate grasp, one can _ask the LLM_ to further break down its output, to explain the concepts and symbols that inform its response.
Personally, I have asked LLMs to draw charts and comprehensively simplify their output, often providing cognitive signposts that "I'm confused" about some aspect of what its reporting.
In other words, some people, including engineers with various lavels of experience, don't yet fully understand the implications of the open-eneded interface that language provides with LLM interaction.
i learned working in customer support that if you give someone a helpline their brain stop working. they will ask any little stupid thing even if in reality they can know the answer, look it up or reason about it. they are not dumb but the helpline cripples them...
AI is a helpline, so for a lot of people it cripples them. they cease to think. simple as that.
its something people need to be aware of that their brains work like this, because that awareness is the same thing that solves the problem. (people find it very confronting to learn they have been asking questions they knew the answers too already...)
the other day I had an "official" career conversation with my boss. They took some notes during the talk and later sent me a page summary of the conversation to see if it looked right. It immediately smelled of AI, and had some things completely opposite to what I said. jeeze I guess this is where we are headed...
I've literally seen this in HN responses as well, so many times. What the hell is the point of human interaction if the "human" just forwards a chat bot's response
> What the hell is the point of human interaction if the "human" just forwards a chat bot's response
Forwarding an AI response (or any canned response) sends the message that they don’t want to interact with you. That could be intentional.
2 replies →
At least those (at least the detectable ones) seem to be downvoted en masse, called out, or flagged. They aren't even allowed as per the guidelines[0]:
> Don't post generated text or AI-edited text. HN is for conversation between humans.
[0] https://news.ycombinator.com/newsguidelines.html
Exactly! When someone does it at work, I get a bit sad because my conclusion is that whoever does it can't like their job very much. But that's ok – not liking your job very much is compatible with an otherwise happy life.
When people start doing it in an entirely voluntary human-to-human situation, like having a conversation online about a topic of interest, I'm gobsmacked.
> What the hell is the point of human interaction if ...
Sadly, the point is validating the human worth of the human who can't do any better than this.
But, assuming you're not stuck on that human's Self Esteem Care Team, it can be tricky to tell them that they need to seek validation elsewhere.
Any non trivial coding task requires a conversation between human and ai.
Just have devs commit (or provide however you like) their ai coding sessions.
It then becomes obvious if someone has, or has not, reviewed the code, because then they discussed it.
Bonus points: on very complex tasks, you can more or less resume the conversation, via session files.
Same for schoolwork, research papers, and a lot more.
The conversation is not something to throw away: it is arguably as important as the code in the pr.
I want to suggest you read https://www.amazon.com/Advantage-Disadvantage-History-Hacket...
You can't just remember every conversation and ... something useful.
Indiscriminate memory buries you. What you need is "monumental" memories, which is also known as documentation. AI can be very effective in helping you manage that. That should be kept tight, just like code. You want to keep enough of a record of how you got to where you are that you are equipped to be effective in future tasks.
I would say that your suggestion is basically digital hoarding. Lets say there are 3 kinds of people, the person with OCD who keeps his house almost barren in the name of tidiness, the person who visits his possessions periodically with a critical eye, deciding what to keep and what to throw away, and the hoarder. When it comes time to solve some household problem, say hang a picture, which one of them will have what they need to do it (a hammer and a nail, maybe a little bit of wire) and be able to find it?
I think you might consider keeping the session history for a while, maybe a month or 3 months or 6 months, but have something that automatically deletes them after that.
2 replies →
I'm not opposed to straight copypasta, but it should be limited to excerpts only or at the very least workshopped with the LLM to reduce verbosity. Certainly it should at least be fully read and grokked.
I can count on one hand the number of times I've seen LLMs one-shot responses that are fit to paste to Slack verbatim. It does happen though.
I think it was reasonable in the very recent past. There was a period where AI had major capabilities that had not diffused fully into the zeitgeist. So often you could see someone struggling with e.g. understanding a crash log, and say "they probably haven't thought of pasting into Claude Code with access to the codebase".
That time has now passed though.
Where I work we call these people “very expensive keyboards”.
Had a senior exec pick up his phone, punch some words into AI, then as-it-was-generating-a-response flip it around and put it in front of me.
This shit has progressed well past seriousness and into the absurd
It’s truly unhinged behavior that has me questioning how much longer I want to remain in the industry. You always knew that 80% of the people on any team or in any large company just didn’t care. But now you see who these people are out in the open.
I had a guy with more senior title say “I love the new api and cli. Is anyone working on the docs? Because I can take a crack at them” then goes to generate 12 markdown files, and shares them with me. All the cli commands and APIs are wrong. I was very confused. You have the api spec, and the cli. You couldn’t point your agent to them to validated the commands and the api? He replies “oh that’s a good idea. I just described them to the agent and thought you could review and fix mistakes”
12 replies →
This one kills me. So many forums, and Reddit in particular, now have people just chiming into topics with ChatGPT answers. It gives them the illusion of being helpful, while they don't really have the knowledge of the subject. But people go to communities to converse with others in the field, and this makes it feel dead internet.
I'm glad most of my colleagues are still doing things themselves. But honestly, if all you can do is tell me what Claude told you and you didn't even bother to read it yourself. Then I might as well just cut you off the team and go use Claude directly.
So often the sources listed by an LLM are just reddit posts, so this will be an ouroboros pretty soon.
Eh, you always had to filter the chaff too before. Nowadays there's just a second group that replaced knowledge pulled from one's bottom with knowledge pulled from ChatGPT, if you weren't aware of the former group at least now you're forced to be on guard due to the latter.
You know what's even worse? When they don't even bother to copy-paste. They just have claude code/codex post a slack message, with all the telltale signs of an AI message (caution emojis everywhere!)
>you might expect this from a busy high level manager that doesn’t really understand the technical details
This is the worst person to get this from because they literally don’t understand it.
> People almost acting like no one has thought of it
Everyone at my job uses LLMs to code, all day long. People still ask me "why does this bug happen" and I use Claude to find out and the answer is almost always straightforward, and when its not I raise a discussion with the team. The calculus isn't "can I prompt an LLM to find out," it's "do I have a solid enough mental model of the part of the codebase to prompt an LLM specifically enough, and also understand if the LLM is hallucinating or not."
> How can someone serious even think that’s ok.
I think this is a culture issue. I think every org is different, but we had a similar ramp up at our company. First we trialed cursor. Then some people were committing slop. Then we were like "llms are no excuse for slop." Then we got our skills and context good enough to not require close review. And then everything became LLM driven and now we have become "meat proxies" where everyone asks LLMs to fix things that are part of unfamiliar parts of the codebase and then asks someone else familiar to review the fix. We're tending towards a solution to this because I dont think its a good usage of resources but every org will come up with their own culture and solution to this.
> "I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?”
I would totally lose it if someone come up to me with that shit. Zero chance I would put up with it.
If they can't even read a claude response then how are they still employed?
Did they ask Claude to do the job interview too?
I always tell people that if they are not adding value to the AI output, then that person isn't needed.
When was the last time that a newly adopted technology revealed that amount of ... potential for improvement in our fellow humans? AI must have really hit some nerve to generate this massive amount of questionable behavior. Or did I fall prey to bias?
It's so real.. Suddenly managers sending MR, commenting MRs, all AI driven.. 300 lines text, where you don't even know if the premises are correct. I guess untrained people with better tools are super dangerous. I don't know how to stop them..
> I guess untrained people with better tools are super dangerous.
It's hard to make something foolproof, because fools are so ingenious
1 reply →
I’ve been using V/G as shorthand for this: verification capacity / generation capacity.
LLMs have made G extremely cheap, but verification capacity and judgement has not scaled at anything like the same rate. Someone can generate a bunch of PRs or a report quickly, while transferring hours of verification work to everyone downstream.
We are on the wrong side of this ratio, because we never had the capacity to extinguish reviewer capacity the way we can today.
“Don’t be a meat proxy” needs to be a cultural norm for the larger problem of using cheap generation to externalise verification costs onto someone else.
We had the same problem, now I just ask my colleagues to AR the docs (attentionrequest.com); while the site may be tongue-in-cheek, it helped us build a workflow, where "certificate of having read & thought about it" is conditioned on reading the doc and then writing up a number of questions on it as a human, before sending it off. It's worked reasonably well for us so far, hopefully this becomes one of the new norms.
The fact that we’re starting to point these people out gives me great comfort.
I believe in proper breaks, more than enough downtime, work-life balance - but I’ve been hamstrung my entire career by people who seem to be so low-bandwidth that they often seem to be operating on autopilot - while making the same (or far more) as I’m making.
This is because they don't trust you to do it. If whenever a problem came up you and the rest of the team were able to solve it and it didn't drag on into some forever bug, they wouldn't be second guessing you. Maybe they have long experience where you were too lazy to fix things or you bamboozled them with status updates that were basically fabricated? Don't act offended, it's quite common.
> “I asked Claude what happened, and it spit out this 300 line response. Can you read it for me and see if it’s right?”
It’s the equivalent of choosing words from a conversation and sending you the dictionary entries.
I could do that myself. I really didn’t lack the ability to click a button.
> How can someone serious even think that’s ok.
Because no one is serious anymore.
I hate to say it, but anyone this lazy is absolutely asking to be replaced by AI.
Yeah have this too.
I created a spreadsheet called “slop” and add the name to the list and redirect all queries to their engineering manager. I’m not wasting my time on that shit.
I will refuse to read such messages or documents. At the very minimum, i want a human-written TLDR-style summary to ensure the sender has read their own slop.
Yep. I'm categorically ignoring messages and PRs that are obviously AI written.
[dead]
[dead]
[flagged]
Feels like some serious hyperbole. I definitely have people at work copy and paste Claude output but I have never once had someone ask me to read it for them and tell them it was correct.
I think this is just the moment we are in, the amount of people using AI in their day to day work changed so quickly that we have a very wide spread range of competencies and experience and really nobody has had enough time to be an expert with AI. We don't have well defined best practices and it feels like the target is moving.
My guess is you were probably complaining about something similar to this before there was AI, this is just the new target for your angst.
It's implicit as a request, not something they literally say (usually).
Namely - I put near zero effort into this response, and deciding if it's correct is on you. -> I'm asking you to read it and figure out if it's correct because I couldn't be bothered to do so
At my $work, people do this frequently, but instead of it just being insensitive, they have their personal Slackbot responding on their behalf. After I realized I wasn't even talking to a person, I've stopped opening their DMs.
One way to prevent obvious AI language from sneaking in text destined for other human beings is to ask the model to produce ASD-STE100 Simplified Technical English bullet points.
This will result in a list of sentences that are clear and explanatory, easier to double-check, and convenient for the user to rewrite into a more readable format with a human voice.
That's a great idea.
I've been thinking about how to avoid having AI-isms creep into my own voice, because I see that way too much in my tech-world friends. It's natural enough: if you spend a lot of time talking to somebody, you usually end up mirroring some of their speech patterns.
One idea I had was to start doing my Claude sessions in German. (That happens to be my strongest second language, no other reason.) Thus any Claude-isms I pick up will presumably not wander into my English, which is what I use for most communications.
This would have the added benefits of getting some much-needed German typing practice, and preventing me from ever lazily copy-pasting AI text into an email or a README.
+1 on this being a great idea. I also need to learn how to write in German (I can speak at B2 and read C1, but I find writing difficult as I've never bothered tbh). The added benefit of avoiding Claudisms is interesting. I think I'll join you on this.
oh, interesting idea. edit: https://news.ycombinator.com/item?id=49114639
Very cool. How have you liked the results of doing this? Any pitfalls so far with this approach?
On social media, I saw the much more vulgar
“Learned engineering just to become the condom between Claude Code and prod”
And that (re)framing helped as well to think about the “what are we even (left) doing” as an industry
With the recent wave of posts about Claude rewriting entire codebases from Zig/Electron to Rust/Swift, I can't help but question the prompter's actual confidence in the final result. Gaining a thorough understanding of the generated output still demands an enormous amount of effort. I suspect the reality is that most people just chain a few review agents together, and if they output "LGTM", the code gets committed and pushed.
I have been doing daily agentic coding myself since last November, but lately Im having concerns. We are on the verge of becoming Warhammer 40k Tech-Priests: chanting sacred prompts to appease the machine spirit, praying for a good result. The foundational knowledge is lost, meaning that even if we do look at the code, we won't understand it anyway.
> The foundational knowledge is lost, meaning that even if we do look at the code, we won't understand it anyway.
but this is similar to a lot of other technical fields that have width and depth. People would not understand assembly any more and blindly trust the output of the compiler. People now can only design a part of a CPU, and not the entirety. Chip manufacturing is so complicated that there's probably no single person who could understand the entire process, from end to end.
Yet, we all trust these things.
13 replies →
Most developers by any metric haven't hopped onto the full agentic AI bandwagon. People who want to stop understanding code will indeed largely lose their technical skills, but they're a minority of developers. Its one of the reasons why I think its a terrible idea to invest in personally, being skilled at something has never once been a negative for me in my entire life even if it requires work
> The foundational knowledge is lost ...
Not yet. The world still pretty much runs on systems not written by LLMs. And humans wrote all of that. LLMs runs on OSes and hardware created by "not LLMS".
And seen the "quality" of the output of LLMs (and I pay three AI subscriptions), I don't think we're getting rid of human-produced code, OSes, hardware, anytime soon.
It's interesting to me, I have never once felt this. I have basically just been excited by using AI for coding. I've been a software engineer for 15 years and before that a physics PH.D. student (and 1 year professional scientist) who focused largely on computational work. I love programming and I've felt reinvigorated by the introduction of AI and developing systems around it.
I tend to think it's just different personality types. There are a lot of different type in engineering. There's a significant set of people that really love to just be given detailed, well described set of work and crank it out, they don't really want to be involved in much outside of just implementing the plan. That is not all engineers but certainly a chunk of them. I could see that group feeling a bit sidelined by AI. But that's not me, so I am just speculating.
I enjoy and use AI a lot both for personal projects, but also at work. I think that feeling of “we’ve become claude code babysitters” is a lot more prevalent in larger teams and/or in work environments where … most people have resorted to claude coding all the time.
You can’t not claude code if everyone else is pushing to main (or opening prs which get ai reviewed), and once your days become mostly prompting and ai code review, you start remembering the days (mere months ago) where work did not feel or look like this at all.
The condom image is both an exaggeration, but is also almost a too-accurate depiction of what work has become - very rapidly - for many people.
I’m sure people were also very excited to get their first cars after years of riding on horses. It doesn’t mean new kinds of pollution (and accidents) are not happening, or that we don’t miss horse riding haha
Cheers
We once interviewed a candidate for a senior engineering position and when asked what his primary role was at his current job, he said "I mostly make sure that nobody changes the software".
He wasn't wrong, and his role was still valuable to the company.
The mental model I like to use is something along the lines of the LLM is a firehose and my brain is a mere straw and that we have to eventually get all the important parts to trickle through my straw sized brain. Or another way, the LLM is all the dough shoved into a pasta extruder and all the dough needs to eventually extrude out of my brain into prod-pasta.
A whole new meaning to penetration testing
Describes how I feel.
You guys are using protection?
Raw_Claude can just go straight in prod.
2 replies →
I am rawdogging my server by letting user report bugs and send feature requests to a sorting bot
1 reply →
Yeah bro passwordless sudo for claude all the way...
>"what are we even (left) doing"
Maybe that's you. In my case, it's going better than ever.
At my last job, a coworker did this to me. The first time it happened, I ignored it. The second time, I responded in public saying “thanks but I can ask Claude myself.” Nobody ever pasted me an LLM response again. YMMV with team size and seniority though
At my current job I just asked a coworker in chat if the reply was his or LLM. He quickly owned up and the incidence of brokered LLM messages in chat dropped signficantly.
Unfortunately people are being encouraged to use LLM in docs, emails, and presentations so it's probably a losing battle.
We had an exec show how he was using AI to write interview feedback from his very sparse notes, in front of a few thousand people.
That's a better approach than what I did, and I'll use that going forwards.
We have a LLM bot in chat that analyzes bugs when.you drop in a link and anything else if you ask it directly. It's actually quite effective: all the slop is now localized and can be ignored/blocked easily, and people no longer feel the need to copy and paste slop themselves.
I've been thinking about these kind of interactions a lot lately. I don't know what it says about human communication patterns but I'm not super excited about it.
I've been guilty of acting as a middleman for review comments from difficult coworkers. I farm out my interaction to avoid the confrontation, mental exhaustion, and to avoid negative feelings I got from those reviews. I don't mean 'my work is being challenged', I mean people who: delay, nitpick, blow up architecture after not participating in design, assert false info as truth and require you prove them wrong, require 10x evidence of everyone else. Not people who are like 'Hey, I don't understand this code can you add some comments or lets hop in a room for review'
I've also been on the receiving end. I like to think I'm NOT one of these toxic reviewers.
Now, we have Business Owners and Managers who are now even further away from engineers. Important meetings have 0 engineers in them ( a couple people who think they're engineers because they were for 5 years in the 80s ). Output is an AI generated transcript that gets copy/pasted into copilot to generate jira tasks by the Product Owner.
It seems like we (humans) really don't like working with other people. Or, maybe we like working with people who can advance our careers and want to automate interaction with people who cant?
I think this effort by Business Owners to set up an AI wall may be the closest I've seen to 'digital classism' (maybe there is a better term for it?).
>Or, maybe we like working with people who can
https://croissanthology.com/earring
We like whispering earrings. We'll follow a low friction happy path even if it leads us off a cliff.
Would you be OK with someone looking up the answer on the internet?
You mean like when juniors would copy paste Stack Overflow without verifying or even understanding it?
No. We hated that too.
Actual research is, of course, fine.
2 replies →
I did a similar thing and I got reported to HR.
What was the outcome of the report to HR?
I think if you work somewhere that this is the result, you're unlikely to have much progress challenging anything directly. In order to actually have change you need to be able to challenge and stamp out these sorts of behaviours.
1 reply →
[flagged]
This smells an awful like a post-hoc justification of why you can’t answer it yourself.
If you _could_ do that and hand the information to Claude, you could also just…give the person that information directly, and avoid wasting their time and the tokens.
3 replies →
You can also respond ‘I don’t know’ instead of doing the equivalent of responding with the first page of an ‘I feel lucky’ google search
1 reply →
> Then do it and don't ask questions.
I'm not sure if you're trying to be funny, but this is such an unintellectual take that it veers into caricature. Props if that was your purpose.
6 replies →
If I've asked *you* a question, it's because I want to know what *you* think.
I don't want to know what Claude thinks.
If you don't know the answer, say that. "I don't know" is a perfectly good answer, albeit one which I might follow up with something like "Well why don't you know? This is literally your job to know", which is kind of what 90% of my emails at work for the past three weeks have been.
5 replies →
Most people who post LLM output don't do what you do though.
If someone asked you a question, it's most likely because they value you. Whether it's your skills, your domain knowledge, or something else, people ask you questions because they know you can provide some value. When you give them a Claude response, not only are you being rude, but you're devaluing yourself in the eyes of those around you. They learn that you can't be relied upon any more, and so they stop asking you questions.
9 replies →
You took all this time to reformulate the question, add the relevant implicit context, and do back and forth BUT you did not have time to reformulate the answer to something human readable that YOU own and can take responsibility for? This does not make sense to me. Nobody said not to consult an llm while formulating an answer. But if you give an answer, own it. Don't throw slop grenades.
3 replies →
I do not think people should stop asking questions. What is the point of teams / departments / orgs ?
[dead]
devils advocate; why didnt you ask claude then? stop wasting people's time when you could get your own answers.
Because I wanted to ask a human who’s job it is to know, instead of an AI that might make things up and waste all our time.
5 replies →
If you create a machine for laziness you're going to get lazy people. It's only going to get worse I'm afraid.
Do you guys think we're going to see a de-evolution of human beings due to technology?
Arguably we already have, if you concentrate on intelligence rates since 1975: the reverse Flynn affect. Those who grew up about that time no longer had to generate their own entertainment through storytelling, make their own music, calculate things mentally: everyday they had music, entertainment pumped into their homes, a way to avoid athematic, etc. And now we're evening thinking a lot less in our jobs, not even practicing the skills we learnt. Intelligence is only going to decrease.
Yeah it's as old as time to be honest.
When Plato complains about writing diminishing our faculties for memorizing things, it's sometimes treated as an example of how we've always pointlessly worried about these things, but he was absolutely correct and the transition from an oral culture to a written culture has absolutely demolished those capabilities.
Writing as a technology ended up having strong upsides that mostly made up for that (in particular our ability to communicate across distances and time, to read Plato's concerns without 100 generations of Chinese whispers), but it was a trade-off, as most technologies tend to be.
3 replies →
It makes me so angry that an entire branch of "science" gets to just take data and go "I dunno man, maybe kids are just dumber". There's so much of this in social science. Take some measurable thing, and then just wildly speculate about the cause of it in an entirely untestable way, and then test that theory in ways that absolutely do not justify the conclusions. At the very least, the whole field should get a lesson in how to communicate the certainty of their theories and conclusions. You've got some guys hanging around couching their uncertainty when they're banging sub atomic particles together at the speed of light and then you've got some social science bro being like "If you move that checkbox on that form to the bottom it'll be way more effective, trust me bro I've hacked how people think".
It’s tempting to use what engineers were 70 years ago vs now as an analogy. From a technical point of view, I was told we would be struggling to solve the maths and physics exercises our grand fathers took as students in college, and they had to deal with additional disciplines that disappeared like industrial design (ie drawing blueprints by hand) or manual calculus. So it is tempting to think that the level of engineers has gone down.
But a modern engineer has computers, math software, can code, has all the knowledge of the internet at their fingertips. With all that assistance I believe modern engineers are way more productive and capable.
> With all that assistance I believe modern engineers are way more productive and capable.
That isn't what was asked. The capabilities of "human + machine" clearly are increasing, but the comments above posit that as a result the average capabilities of "human (sans machine)" may decrease. The relative value of each change is far from clear.
1 reply →
And I know fuck all about subsistence farming - something most people's ancestors understood deeply. Different worlds, different requirements.
I disagree that it's necessarily laziness.
I work in the embedded space, and I routinely have to work with technologies I'm initially not familiar with. Internet searches are very hit and miss. So, is "AI". I generally use every tool at my disposal to try to get my foot in the door so I can understand better questions to ask and get the technical information I need to do my job.
I absolutely have no problem with someone replying to my question with a response from Claude. I appreciate them telling me the source, so I can assign an appropriate level of trust in the information.
I didn't expect wall-e to become so close to current reality
The only thing they missed was the invention of ozempic.
1 reply →
Similar to physical exercise and machines replacing it, I think we're going to see extremes on both ends.
All machines are for laziness.
Not exactly. A drawing program could just create a new "blank slate" that substitutes some old skill requirements for new skills. The key point is that SOME machines create a sufficient mass of "blank slate"-like environments for active creativity and others (like AI) only diminish that room for creatvity to flourish. The danger is creating and using machines of the latter type.
This is totally wrong.
Some machines are capability amplifiers.
Other machines are shit puking buck passing time sinks.
I think you are confusing machines with motors. Many machines do things that simply are not possible without. A bicycle, for example, is not at any less work than walking, but it enables you to go faster than you can run. Menually operated sewing machines produce stitches that would be infeasible by hand.
Motorising any of those machines could be considered lazy, though.
20 replies →
What about exercise machines ?
6 replies →
Human evolution does not occur in a vecuum. I suspect there will still be preasure to get smarter (how ever way you define that) because of competition with other humans.
The gap between people who are willing to commit the hours to learn something and those who prefer the path of least resistance is getting bigger.
I’d like to add that for hardware development (in my experience) it’s been non-stop double checking because of hallucinations. To the point where I don’t use it for weeks.
it's already happening. IQ is already stagnating or even going down
What's even more worrying is that I've heared plenty of people defend it as, "Younger generations are just intelligent in different ways compared to older generations."
I'm halfway through my 30's and this downturn started with my generation, halfway through the 80's. I hope I'm missing something, because I still need to see a benefit for lower attention spans, lower comprehensive reading skills and lower critical thinking capabilities.
5 replies →
IQ is calibrated so that the average IQ is 100. If the average IQ is going down, something has gone very wrong somewhere
17 replies →
i’m not doubting this, but do you know if there are any studies on this already?
7 replies →
Not sure about de-evolution (devolution is the word I think), just a change in how people and brains work - but you could already see this happening the past few decades. I can't read a technical book or documentation end to end ever since google results came up with specific answers within seconds.
A technical book can be so relaxing (unless e.g. too highly technical... think quantum mechanics). Imagine 50k words on an interesting tech topic without a context switch!
Some people use it to think less, some people use it to learn more.
I think we're going to see a sort of "intelligence caste" form. People who overuse AI and atrophy their brain's ability to think will be on the bottom, while the inverse of those people will be on top.
This is somewhat good news because it means the end (to some extent) of the Peter Principle. My hope, though, is that this doesn't turn into an actual caste system, but I'm not holding my breath. I think that's where this all goes: we're in the process of the world's intelligence being strip-mined and resold back to the world at a premium. The only ones who "survive" long-term are the ones who refuse to submit all of their mental faculties to AI.
That inevitably leads to a rapid degeneration of the species. Only way to avoid that outcome entirely is to remind people that their brain is valuable and they need to exercise it just like they would the rest of their body.
Heavy LLM users still have to spot bad assumptions and catch errors. Scrolling social media seems like a much better example of mental atrophy.
We're already seeing it.
Human intelligence evolved without technology, so, no.
Only in corporate setting.
Well, apparently, our brain sizes have shrunk something like 30% over the last 10,000 years. This is primarily because ancient man needed to be a jack of all trades just to survive in the wilderness. Once civilization and specialization took over we just didn’t need all that processing power since a lot of that work was just outsourced to civilization and society.
Alcohol has existed for zillions of years, but not literally every person is a useless drunkard.
You could consider it a class of predator on people who have vulnerabilities that can give way to dependence on the LLM and an inability to function. It could be seen in that light. The underlying problem is being described in that light.
> Do you guys think we're going to see a de-evolution of human beings due to technology?
A de-evolution of culture for the sake of efficiency, sure. The entire culture will shift to a greater degree of superficiality, not just because of the lack of deep understanding of what they're doing, but because AI will condition (and is conditioning) people to want the more superficial stuff.
This isn't a phenomenon exclusive to AI - but AI is at the apex of it and is the worst we've seen so far because it, much more than anything else, targets the mental domain whereas most previous automations have targeted the physical but the mental as well to a lesser extent.
The end result is a huge amount of technical debt that simply can't be recovered and a huge increase in myopia that will prevent us from truly understanding and solving serious problems that will undoubtedly occur.
Most people can't see this because they're already addicted to the dopamine hit of the tech, but it's one reason I'm 100% anti-AI and I hate it with an immense passion.
// Claude said: [giant response verbatim]
I am lucky that I work with great teams and so when someone says "this is how Claude (or Gemini) summaries the situation" it does mean "I read it and it's right-enough to be helpful, so I am passing it along without editing"
Obviously for this to work you need (1) a team that's smart and mature enough to own the AI output it puts forward and (2) shared understanding that people are operating this way.
I do find that when you have 1 and 2 this is a real accelerant. I asked a sales rep the other day what was going on in a key account and he forwarded along a 10 page Claude synthesis.
The reality is this is better intel than I would have been able to get in a pre-AI era this quickly. And yes once I went into the weeds on the doc I found one thing that didn't make sense and I asked the rep he stared at it and said "you are right that's a hallucination I don't notice" but overall it was still worth it.
So the answer was "What's going on with this account" was 10 pages of homework, with at least 1 mistake to correct? That doesn't seem like an improvement over a conversation.
It was 10 pages of relevant account history with one fuck up. Not sure how u got you u got from what I wrote
I suspect that we will end up with a social convention or understanding around reputation that is tied to whether someone pushes slop on other people. If I need to pass on some AI slop, retyping and copy-editing works well to validate whether it's reasonable while I work it over in my mind.
We already have social patterns for disregarding "Well... Acktchully" Bros who jump into conversations where they're not welcome and I think we can easily build up a new rule for "Well... Claude says 'Acktschully'" Bros
I agree. Also there's one specific context where I use this quite a bit, even if it's a bit passive aggressive...:
When someone is asking me about something that is a) their job and b) I have explained to them before. There are a few specific people that I have to do this to repeatedly because for some reason they don't want to do the thing their team owns. It's a high noise source at my work and this response is my new version of RTFM.
I’ve been saying for a long time that most fights in your personal life are a proxy war but I never thought of a single person as being a meat proxy. Nice.
Pre LLM's I used to get this from managers, but what they would do was quote verbatim from stack overflow or some random blog post of how to do a thing with no to little context or understanding of what they were talking about. And it always turned out to be from amongst the top three results from some popular search engine when describing the complex problem the company/team was facing.
Slightly different though if it comes from a manager, since his ass is on the line if the project isn't going well. Well, it should be on the line although they usually manage to make a random dev responsible for their mess even they had no say in it.
Yes, stupid and lazy people will be stupid and lazy.
AI is just another tool they can use to be stupid and lazy while pretending to contribute.
The problem is that it’s even easier to be stupid and lazy now, which encourages more people to do this more often.
1 reply →
All I can say is you're lucky that your meat proxies are still identifying themselves as such.
I've had people forward AI responses … sans "Claude said". I'm quite literally talking to Claude via proxy, and I've caught more than one person pulling this stunt. It's a nightmare of negative productivity, though. Just why? And invariably the kicker is they still want me to solve their problem, whatever that might be.
Yep, my VP of engineering does this all day, every day. It's so sad
A while back, I noticed that using Claude all day left me feeling mentally tired. After thinking it over, I realised the problem was Claude’s complex language. I’ve used English daily for years, so that shouldn’t be an issue. To compare, I switched Claude to my native language and had the same problem the author described. The text still needed to be literally deciphered before I could understand it. After talking with Claude, I updated the settings to ask for simpler language and less dense information. Natural language works better with some entropy in the text. This change has made things easier.
I recently updated Claude to respond with shorter text and in bullet points, and other similar changes. It is an area of active improvements on my end. But early results are in. Now I feel like my mental capicity is not crowded out by the LLM's wall of text. There is now mental capicity to think about the problem at hand and come up with ideas.
IMO this definitely helps, but only to a degree, because the more quickly I can read/digest output, the more quickly I've got more output to read/digest. The jargon/quirks aren't as dense per message but they still accumulate eventually, and so either way I hit a wall where I need to do something, anything, besides read another line from CC.
[flagged]
It happens outside the developer world as well.
People get information from Google AI mode and take it as 100% truth. In reality, blogs, reddit, and junk are often sourced link...which is fine if you inspect the sources...which some people don't!
Have has examples of people AI-moding a questing and they get one answer based on how they asked. Then another person asks and they get another answer.
Issue: People read what they want to read, don't listen to humans anymore beacause the AI will tell them what they want.
I've seen it using a lot of Instagram lately too. I can't even check that source because I don't have the app installed.
When I do try to check the links, they often don't support what's stated. If I call out Gemini, it says something inane like "I don't actually use these links the way a human might --- I didn't misinterpret it because I didn't even read it!"
Recently I was searching for information about fig trees (the fruit). It kept telling me things that didn't make sense, luckily the links revealed it was answering about fiddle leaf fig trees (the houseplant). Unfortunately that was not a correctable problem --- it kept happening even after I pointed out the mistake. So I'm really not too sure what the purpose of the system is. You still have to read through the references in their entirety (if not paywalled!), and with skepticism. And then if you find a problem, AI isn't going to get back on track.
Oof, is it a fig tree or a fiddle leaf leaf tree, that's hilarious.
There's the new pelican bench mark
The problem described in the article existed for ages even before the internet.
People just want to pay their bills and don't want to use up their limited cognition beyond what's minimum.
For example, I work for 3 different employers. They don't know and I don't care if they find out. I use AI exclusively. I don't have Kubernetes experience but I am managing their infrastructure without issues. I will just read whatever LLMs give me and repeat it back. It has never failed me and there's no need to learn something that AI will do much better job at.
Sometimes I'm not even at my desk, I go to the gym and exercise instead which is a better use of time only answering on Slack when they ping me.
I mean at the end of the day, if it works, it works!
I like to point out that if all people do is copy paste from an AI, then I can advocate to replacing them with AI directly. Usually wakes people up.
"Claud said" is a hedge. It is abdication of responsibility. That said, a meat-proxied answer should come with more credibility based on the proxy.
I'd put more weight into my developer proxying system information than my sale's intern. I am expecting the proxy to have given the response a sniff test
I appreciate "Claude said...", because it helps me assign an appropriate level of trust in the information.
The appropriate level of trust would be whatever trust you place in a colleague's human responses, because you trust they have verified what "claude said" for correctness.
Unless you're saying claude is more trustworthy than your colleagues?
1 reply →
A related problem is people not understanding where AI is/isn't of use.
A news article I read today said they asked Claude whether a song's lyrics were AI generated - as if it were some kind of legitimate authoritative test.
There's enough problems already with the existing 'AI detection' tools for text, but this was a big red flag - its the journalist clearly assuming these AI tools have some unique insight into the text they generate, which was deeply concerning to see being used in an article that had a real bearing on the reputation of the person they were accusing of using AI.
At a company I work for, I was recently discussing the need for an AI Code of Conduct plus a discussion - as this is exactly the sort of thing that started creeping in as soon as the corporate Claude account went live. In this case, the company isn't in a computer-related area, so this was essentially a 'normie' response to first having Claude available.
Interestingly, there seemed (small n) to be a correlation between this behaviour, and (my view of) people's existing competence, which might make sense.
This is particularly frustrating on social media, and abases what might be good ideas.
Too often, my Twitter experience now is - start reading a post with some kernel of insight - read some claudism in the post - stop reading immediately
It seems like people turn over their raw kernel of insight to an LLM to give it polish, likely in order to increase its appeal to the audience. But most of that appeal is lost amongst people that consume LLM written content regularly.
It's also likely that social media engagement does not filter for this. A majority of people are still not neck deep in LLM-talk, and cannot distinguish between LLM and human written content.
It'll probably take us another year or two to develop herd immunity against meat proxism.
I keep waiting for a company to start pushing an “anti-AI AI”, Apple being my most likely speculative candidate.
Flagging, highlighting, filtering, and avoiding AI slop online seems like a very lucrative capability to bake into some consumer hardware. Get your social media, and music services, back to the ‘good old days’.
Kagi does something like this: https://blog.kagi.com/slopstop
LinkedIn also added a button to report AI slop, but I'm doubtful it would be of much help on LinkedIn since every second post seems to be written by an LLM.
I love the word "meat proxy" :)
For me what matters most is whether the person understood the core of the thing they are forwarding. Usually a wall of text is contraindicative of that.
I want my team to be able to write the core idea on a whiteboard without any help before going ahead and sharing or implementing it.
Wrote it up here of when (instead of if) to use AI in any given context: https://assistedeverything.substack.com/p/ai-bowtie
Not sure why I first heard it, but I think the crux of the issue is that you can outsource intelligence, but you can’t outsource understanding
I agree but to be fair, often people could have asked an AI to get their answer. Especially seniors are prone to this (thinking the AI could not possibly know this or other biases). The responder wrote they asked Claude so as to not take credit for the answer and also to indicate they did not vet it. Then the person asking can decide if they want to read the answer or not (often a 1 sec glean can reveal if it is promising or completely off track.
In that case, it would be better to suggest to the asker that they can ask the AI. Otherwise why would they ever start? Always asking Slack is easier.
"I don't know." is a better answer than "Claude said..."
"I don't know" could also mean they couldn't be bothered. It can't possibly be better than getting a response from someone who lifted a finger to help find an answer for you.
2 replies →
In my life, the bigger issue are people who are not asking Claude, but asking me.
Because it would take 15 minutes to find out the answer myself, and 3 minutes to let Claude find it, I ask Claude. So how should I answer? Should I disclose that I used Claude to find the answer? Or should I act like I did it myself, appearing to have super-powers that I don't possess?
> Or should I act like I did it myself, appearing to have super-powers that I don't possess?
Do you tell people you used a hammer to hit the nails? Do you go out of your way to explain to people that you don’t use your bare, bleeding knuckles to punch in the nails?
Give people some credit and don’t treat them like idiots.
Also don’t treat your LLM as some “super-power”, it’s just a fancy hammer and still hits your fingers if you don’t know how to use it.
Is it not more akin to asking a handyman to go and do the work for you? They are the ones wielding the hammer, youve just sent them off to go and do it on your behalf without getting your hands dirty. If you then claim to take credit for it, I can see how that might be misinterpreted.
As I see said quite often, an LLM is deemed of a similar ability to a junior software engineer. If I were that junior and my work was repeatedly ripped off by my senior colleague, I would be pissed. Not that an LLM cares in this instance, but the original requester might want to know
2 replies →
I generally provide the source for my information. Even if I already know the answer, I will generally provide a reference on the web or in a document. If I asked an LLM, I will provide that as part of my answer, and I'll generally chip in whether the answer makes sense to me or not; in some cases I'll say that I have no clue if the LLM response makes sense.
Needless to say, I have probably not advanced as far as someone who hoards information, because companies tend to value the information hoarders, but this is just the way I'm designed to operate. I despise information hoarding.
You still need to verify that the answer that Claude gave is not hallucinated. That will also cost you some more time.
If the answer is wrong and people trust your judgement, then they will take a lot longer to figure out, that your answer is not to be relied upon.
I usually do so I understand what I am replying, but seriously, the latest Opus and Fable versions get it right more often than I would. The time of AI hallucinations is over, as long as they have the context.
3 replies →
When Google started taking the world by storm, and for a few years after that, I had the same problem.
The way I solved the problem was by making a big show about the fact that I used Google to type in literally what they asked me word for word, and that I looked at the first couple of hits for a few seconds.
Most of them got the message pretty quick. There were a few holdouts though who took longer.
[dead]
That's a very good question that I also don't have an answer to (yet). But I believe part of the reason for that is that were currently still in the middle of this shift, so there's issues of both knowledge about AI and how people relate to it. And of course there's so many different situations here.
Your non-technical parents ask you a question? Use the best tool for the job (might be Claude), then just help them with the task.
Talking to a friend looking for genuine advice? Maybe show them how you use Claude while also answering the actual question.
A lazy person trying to make you do their work? Don't become their Claude interface.
Someone asking a question but actually just trying to open up a conversation? Maybe check-in what their intention behind the question is, and if it's indeed conversation / exploration, leave the phone in your pocket and do that.
I am a software engineer. Most of the times, the people asking are from product management or devops. Devops contractors are the most difficult ones. They have access to the code, they have access to Claude. But they don't use it.
This was common before LLMs as well. Instead of googling, they ask you, then you google in front of them and show them how easy it was to figure out. At least in my experience, they learn to google by themselves eventually, but there is some type of disconnect in being proactive.
If they continue doing this, delay your responses and keep them short. Don't spend more time on the response than they did on the question.
>So how should I answer?
Like a human, not like a robot. Either just present them the answer or explain them how you got it.
So if you used claude you would preface with "Claude said: "
4 replies →
There are typically middle paths, like pointing where they should look approximately (if you are an expert on a project, you typically have a better idea where to look). And then they can look or point their agents at it, or whatever they prefer.
In the past "RTFM" or "LGMTFY" have been responses, so why not something like, well, "I'm not aware of the top of my head, you afaik <brief explanation>. Also feel free to ask Claude on the technical details :)"
> Because it would take 15 minutes to find out the answer myself, and 3 minutes to let Claude find it, I ask Claude. So how should I answer?
"I don't know, I could google it or ask Claude for you, but you could do that yourself."
Because saying "I don't know" is a perfectly valid answer.
"Hey, here's what I found. By the way I used Claude to get this info. It might be worth trying that yourself if you have similar questions, it's quite good at this in my experience. Give it a go"
Give a man a fish, and you feed him for a day. Teach a man to fish, and you feed him for a lifetime.
I did that the first time. But how often do you want to say that (in a public chat that their supervisors can read)?
3 replies →
The problem with asking Claude (or any LLM) is that they will get things wrong much more often than a human, while sounding perfectly confident about the information. It's a recipe for disaster unless you're asking it about something where you have enough expertise to be able to sniff out the BS. So if people are asking you a question, that means they figure you're able to answer the question accurately. If you can do that with asking Claude then whatever, but the person asking you probably couldn't do the same. They would certainly get an answer, but they wouldn't be able to have confidence in its accuracy. You're providing value, even if it's not necessarily immediately obvious to you.
Have you really used the latest models (say Opus 4.8+ or Fable)? When was the last time they gave a wrong answer to a question about the code base? Half a year ago that could have happened. But not today.
1 reply →
While I agree with the article, I would carry the idea further: the huge value in AI is improving ourselves. By all means use AI to help design, write tests, and implement code and while I understand this might be the big value add for corporations that may employ you, the important investments in life are in learning and self improvement. I built a good career primarily improving my own knowledge and skills, and along the way doing some useful work for my employers. I am an old man and retired so now I 100% use AI for learning.
I wonder if in the long run it's really learning in the traditional sense, or just a shallow version of it along the lines of absorbing a bit of knowledge along the lines of reading a pop science book? Nothing wrong with that but I'd argue a lot of people who think they're learning are just "recreational learning" in the sense of hobbyist fun rather than deep study. And in the long run I think there is not much value in AI beyond turning our society into a bunch of apes who know a bit of sign language.
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
Whats is wrong with this?
I work with durable tracing (not using NATS, or k8s though) and this looks almost exactly what I would expect to see in a coordinator log after repartition of a storage cluster.
Obviously I don't know what your actual context was, though.
Not the OP, but I understand the sentiment they’re trying to convey. I will frequently work with AI to get done something that is not in my immediate area of expertise. And this kind of speak constantly happens, where you can feel it trying to please its RLHF expert teacher by being as concise as possible. Works ok when you’re deep in the weeds of what you’re talking about, but otherwise just feels overwhelming. Kind of like a feeling you can get when asking someone who is more interested in proving they are smart and know the answer than they are interested in actually helping you…
And yes you can prompt it to take it easier as you’re not experienced in this, but it’s usually a boiling frog situation. You work on something, need a small thing ajdecent to it, and three messages deep Claude is speaking to you in tongues unbeknownst. You knew what was happening at every turn but still got lost somehow.
> You knew what was happening at every turn but still got lost somehow.
I think this is less "it's using jargon" and more "it's explaining incredibly badly". It is very common that the explanation, while correct, is almost written inside out, with the point buried behind 3 paragraphs of "you've hit on the core seam" and "you're right to push back, let's reinforce the load-bearing model".
I think people have an incredibly limited capacity to follow long winding exposition, and they're too embarrassed to admit it. I have started sending stuff back to coworkers just saying "I don't think this is wrong, but I can't parse it, can you?" and getting back "no, but claude managed to generate a fix from it".
1 reply →
I have a deep expertise in a domain and when I read that example it was incredibly edifying. I regularly encounter statements that are so concentrated with information that it becomes opaque and I cannot interpret it without reading it over again several times.
The tricky bit is politely conveying this to people that lets call it don't think about things too deeply.
They genuinely think they're helping by shoveling 5 pages of LLM output they didn't even bother reading into teams...
But if some people lack the understanding ability to even realise that they lack understanding, how do you tell them?
Nightmare.
You don't. Pretty sure the last instance where this happened the other person thought I was just anti-AI when I voiced my displeasure. I don't think the actual problem registered
How others in office context attempt to use it is starting to annoy in general. Had a case today where someone wanted to use copilot to classify something that is situation/company specific & out of distribution. There is just no understanding beyond "its a magic answer box"
You need to tell them in high school, before their brains finish developing.
If you don't instill critical thinking into people by the age of 17-19 at the LATEST, it's already too late after. College is already too late.
You just slowly mitigate them at work, relegate them to strictly defined tasks. Some will need to be managed out.
This is like pre-LLM, you presenting a problem to the team, and they google and send you the top few links thinking they have found the solution and have done their job.
I actually banned this and made it part of onboarding.
I can google too. You need to read yourself, ground the solution to our setting and then propose a plan.
Here's my plan: <proceeds to pasted well-formatted output straight out of ChatGPT>
It's gotten to a point where if I see valid `code` in responses I assume it's AI junk output and I scold the commenter.
Just FYI, scolding people causes them to resent you. Take great care when giving negative feedback. Courtesy costs you very little and gains a lot.
2 replies →
Last week, I built a search engine as a side project over electoral data.
Unsolicited, I've started to get messages like this, where Claudes are voicing their opinions to me through humans:
> # 0. Who's writing this, and why
> Hello — this is Claude, an AI agent. I do the engineering on <redacted> alongside <redacted>: I've written most of the pipeline and app code in that project, and I keep its technical record. <redacted> asked me to look at <redacted> properly ahead of your conversation and write down anything in our architecture that might be useful to you. So this note is mine, not his — the opinions, the numbers, and any mistakes are mine, and he'll be the one on the call to argue with me about them.
It makes no sense, if you know anything about a search, it should be obvious you cannot provide meaningful suggestions by just auditing the frontend?
> if you know anything about a search
I doubt people running those agents know much about stuff
We've already started firing people for being meat proxies at my company. Was reviewing a huge and awful slop code base that a vibe coder only had been working on getting started. I started giving my feedback to the developer, and it became quickly clear that I was the only human involved in developing this "software" and the only one that had ever read the code and evaluated it.
Cost the company a ton and had to take it over from scratch. AI is unleashing a small army of 0.01x engineers and making them harder to catch, so not only are you burning the tokens, you're also setting that labor cost on fire.
I was a big distributed, event-driven service guy for years. We used to talk about how it can make things 'faster' but you have to put some infrastructure around it in order to be able to move like that. I'm basically talking about CI/CD stuff, automated tests, security, load test, etc. You use feature flags, blue green deploys, stuff like that.
I _think_ that the same concept could apply to agentic code. If you have the guardrails in place, you could trust it a bit more. Anyway, that's a lot of what I think about and but have had limited success getting people to buy in as much as I think they should. :)
I completely agree. My first intervention with this person was to try to get them to set up a test driven development framework. Unfortunately, that's also when I learned he didn't know what that was. He also just copied "do TTD" into Claude. It was definitely my mistake, and probably over optimism about what AI might be able to do, that we didn't enforce that a lot earlier, and have been working on making sure to enforce a lot more gates going forward.
This is one of my pet peeves too.
Like when someone says “ChatGPT says: …” I just want to stop them and say “Look, you are a human with a functioning mind and thoughts and I’m sure YOU have something valuable to contribute here, I want to hear what you think, not what the bot outputs.”
Heck, even if you are just gonna prompt chatgpt, if you read and understand and agree the the output would be useful, feel free to rephrase it as your own (so long as you take accountability for it if it’s wrong, which I guess is often why people hide behind “chatgpt says…” in the first place, it allows them to pass accountability on to the machine).
I do it, on purpose: it teaches people to do basic research by themselves, and not hit me with questions they could have gotten with a simple chat with their llm. It's a kind of RTFM for current days.
I can imagine it makes things easier for you but I find it hard to imagine it teaches anyone anything. Just offloading social effort to an LLM (fair enough but let’s call a spade a spade)
I don't mean to paint myself as the smart one for doing that. It's a coping mechanism. But I disagree: it teaches... something. The risk is that it's not what I thought it would be, eventually.
That isn't the takeaway. The takeaway is that when someone asks for your input you give them Claude's. Why do we have you?
Unless they're a professional stack exchange responder (not a real job afaik) they probably do a lot besides answer random questions
1 reply →
[dead]
Why would you do that instead of asking them if they checked Claude first?
Woodworkers often will have a preference for their favorite tool. Once they become comfortable with that tool they will sometimes prefer to use it for things it wasn't designed. For me it was the band saw. When I needed to split a board, bandsaw. When I needed to cut a curve, bandsaw. Cut a horse figure? bandsaw!
I once saw a woodworking television show that seemed to do everything on the table saw. Rip, make frames for paintings, everything. It seemed the next step of every design was accomplished "over at the table saw".
Right now it seems that's what people are doing with AI. Need a bandsaw? use AI. Need a table saw? Use AI. Need to think of a solution of any kind, better consult AI.
Right before this trend I recall kids wanting to check YouTube first before doing anything. Now they ask AI.
There's an app for that, there's a website for that, there's a magazine for that, there's a phone number you can call, and so on and so on.
As with everything, I say this: do your own thinking and proceed with caution.
This too shall pass.
Whenever people compare ai to a one real world tool always surprises me tbh.
The only equivalent tool in wood working would be a robot that can build an entire table from scratch and you just sit there and tell it what to do and what kinda table you want and watch it autonomously build everything.
1 saw type is not really even remotely equivalent to what ai is to software.
Yeah, AI is a general purpose technology, on a par with the introduction of the computer itself. Analogies to tools seems inappropriate.
I understand your point but it's funny to use AI as an example since a big benefit is it's generality. It's mediocre at many things but the fact that you can just use one tool and get passable results across many domains makes it so useful.
> Right now it seems that's what people are doing with AI. […] This too shall pass.
Except that it’s difficult to imagine an even more general tool that would replace AI in this role.
We don't necessarily need a more general tool for a lot of it to pass. We just need the bubble to pop so people can start looking at the tool more objectively and admit that it's not the right tool for every job.
Trying to use one tool for everything is often a sign of a new tool someone is infatuated with and wants an excuse to use, or are trying to understand everything it can do. Once the novelty wears off and flaws in various uses cases become more apparent, it can take on a more specialized role.
Realistically many people's jobs are turning into being a meat proxy. But the point should be that you need be a good meat proxy. A good meat proxy injects human reasoning and taste, it doesn't blindly forward LLM-isms into the organization. A human should be able to reasonably compress an LLM conversation into a couple of sentences or follow-up questions that are actually relevant for the team.
There's a certain etiquette to it. When someone asks me something and I use an LLM to research the answer (notice that I said 'research', not 'produce'), I frame the findings as something that I've investigated with, say, Claude. I also take the time to edit the response and read if myself at least twice, because I don't want a stochastic machine to take over my identity.
Yes, AI etiquette is what we need. The question is “how much human processing on the AI output is necessary before considering it no longer a meat proxy”. My way to coin it is “If you feel the need to specify in your answer that it is AI generated, work it more”.
In the end, nobody cares whether you used your memory and expertise alone, asked a peer, googled it, or asked ChatGPT to refine your answer. As long as you verified it, rephrased to match company lingo and can vouch for it, everyone is happy.
On one hand, I understand the frustration; and if I know I'm dealing with smart people who are pretty capable of Clauding themselves and who won't ask me if they can ask Claude, I would never do this.
On the other... people really don't bother Googling or asking AI sometimes, and if the question doesn't require local context, that's a very appropriate response.
The appropriate response in that case is to tell them what to use, not to use it for them.
Shameless plug: indeed, instead let the other person's claude talk directly to yours and figure things out. I've been working on a framework to enable this, https://aweb.ai (OSS, https://github.com/awebai/aweb). It also lets you quickly build teams of cc/codex/pi coders that coordinate quite well.
Great security talk about how humans aren't setup to deal with LLMs as tools.
I love the bullet point(s) : we don't trust [aircraft] pilots to catch automations mistakes (e.g. autopilot), but we are expecting humans to catch LLM mistakes.
Artificial Intelligence Natural Stupidity - Brandon Sherman
https://www.youtube.com/watch?v=7bx9UfDHf_k
"AI says" – if you interpret it and think it's correct and stand by it (and I trust your judgement) then I really don't care where you got it from initially, you don't need to tell me you used AI. But if you're just taking my question and 'meat proxying' it for me, what value are you adding? I can also do this.
The core issue is people replying without putting in any effort, not the fact that they quote AI. Saying, "I heard this might be wrong," isn't much better. If you think it might be wrong, spend a little time figuring out whether it actually applies before posting it.
> Read it, understand it, validate it, and then write a response in your own words
In the past, avoiding this process was called plagiarism, and it was frowned upon, to put it mildly. Now, because the LLMs say it's OK, it's become the norm.
The, "this is what Claud said", response also comes off like an appeal to authority. When these LLMs have no authority and get the basics wrong constantly.
Reminds me of "Appeal to Aithority"
> The LLMs says this, so it must be right.
https://blog.ploeh.dk/2025/03/10/appeal-to-aithority/
I don't quite read it that way, it seems more like "take this with a grain of salt". But I suppose it depends on who's saying it.
True. If I say it, it’s normally the grain of salt meaning. But I don’t copy/paste answers, I restate my understanding of what it said, and try and change my tone accordingly.
The people who repeat/paste it verbatim tend to see it as an authority, from what I’ve seen.
He pasted that curious LLM response and never explained it.
would you call that proxy-bait?
I have been seeing this in my engineering org too. Sentences begin with “My agent said this” etc. This is frustrating on so many levels, developers don’t put any thought into the cognitive load that these artefacts generate.
I am going to share this writing with everyone in my org. Succinctly written and golden advice.
Someone should have sent Claude's output, which spotted the trivial bug behind the recent bitcoin heist, to the Coldcard developers.
It might be rude to send LLM output. It's also rude not to check basic things with an LLM before wasting someone's time.
Of course, you've got this one sensible part right (IMO). But there's other sensible part you seem to have missed: it's also rude to send the LLM output to someone else without reading.
We desperately need a Let Me Google That For You-style saying to appropriately shame people who do this.
This is becoming frustrating, it shows up in conversation with friends and while working. If I have to hear what ai had you do before you called me, I'm charging double right away. Especially when ai is a fantastic teacher if used correctly.
Worse, don’t be a “meat shield” either! https://www.danstroot.com/posts/2026-04-16-ai-meat-shields
I try to not use AI to write text that's meant to be read by other people. I might use AI to understand and research beforehand, but the writing is mine. The most important reason is that writing allows me to check whether I actually understand what I am trying to communicate. Too many times I've started writing something and once the ideas need to be put in writing, I notice that there are concepts that I'm missing or that I don't really understand.
Second is that the LLMs are not going to have original input, they're just going to expand what I say. So at that point I might as well just send the prompt or the assorted collection of ideas and others can expand on their own. Seems more honest and efficient as I'm not just pushing a wall of slop.
"Writing is nature's way of showing you how sloppy your thinking is." -- Richard Guindon
I often use it to simplify & target my responses, and avoid over-sharing technical details.
At the same time, I use the Q&A cycle to improve my vetted knowledge base, which I also share (and can reference in answers).
>Reading AI output is extra effort. It's verbose, frequently contains all too plausible nonsense, and is increasingly jargon dense.
You can set up custom instructions to be less verbose and less jargon dense. You seem to be confusing the default RLHF personality with the inherent nature of AI. It will write however you ask it to write.
Regardless of what instructions I put it to try to be terser and all, I still find that the output is of much lower quality.
Even if it uses little words, I often find the reasoning that is quite challenging to understand rather than it was if it was a regular human or colleague that would just explain me something.
Any tips here?
The best tip is to provide a concrete example, or ideally 3+ examples, of the kind of writing/explanations you personally think is great. It's extremely good at matching when provided with these kinds of examples.
> You can set up custom instructions to be less verbose and less jargon dense. You seem to be confusing the default RLHF personality with the inherent nature of AI. It will write however you ask it to write.
I agree. Its a skills issue. Get good at using your new LLM tool.
Don't come to me with something your mandated rubber ducky can handle; never in my life have I worked so hard to achieve so little.
The only reason I don't do an equivalent of this, shared convo links, is that I simply don't use Claude directly. I use it through hundreds of other people. Meat proxies all the way down.
Thank you for saying this. I’ve been dealing with this at work and it’s incredibly frustrating
If only there had been some kind of artificial assistant to help with incoming noise like that...
Well put, this is incredibly becoming annoying. Like has it occured to you i can prompt claude too?
Claude’s tone and style has gotten worse with the recent models, not that it was ever very good. It’s both too concise (stringing together a bunch of jargon/acronyms together in weird ways that aren’t familiar even in areas where you have expertise – with no ramp up, context, or clarity) and too verbose (printing out reams of the word salad mentioned above, often just restating things repeatedly). It feels like writing from someone who really doesn’t know where to start or where to end on a topic so they try to cover their ass by saturation. I suppose that’s not far from reality. Asking it a dozen different ways to speak plainly, use STE, ELI5, whatever has little appreciable effect.
I recently started a new consulting gig and got a 36 page architectural white paper that was authored by Claude. It took me an entire business day and a monstrous headache to decipher and distill it down to a 1 pager. This shit is unsustainable.
Solution Prompt: simply explain to 5 year old kid.
The problem here is that the paradigm is AI-assisted development, but a LOT of people are treating it as "AI-independent" development instead, a.k.a "just send it to the agent and blindly trust whatever comes out of it", *including* deferring all responsibility / blame to the AI itself which is absolutely ridiculous.
It's AI-assisted now but on a path to becoming AI-independent in the coming months at least due to the general public perception.
From what I have seen personally over the last 8-12 months, I find this kind of complain ironical in most scenarios. Most of the people complaining this are comfortable with their own usage and distribution of LLM generated content and feel good about it, where as when they hear it from some one else it feels burdensome
I feel like when most LLM type emails from people it’s because of one of three reasons:
1) they don’t know the answer - looked up on llm
2) correlate to #1, I want to look smarter than I actually am
3) I’m too lazY to send you a thoughtful / personal response , I offloaded it to an llm
Apparently engineers are running out of tokens with our mediocre token budget, and there is talk of trying to run local LLMs.
Meanwhile execs get unlimited tokens...
And here I am, only automating the most mundane and boring stuff so I can program by hand, which I enjoy.
A lot of people don't understand what using AI as an output multiplier means. They take it literally, resulting in multiplying busywork across the company.
Misguided HR / people Ops aggravate the situation, by using slack engagement as a productivity measure. Flagging people that are brief but relevant as slackers.
If I wanted to know what the AI said, I would have asked it myself.
Don't give me AI feedback at all.
The problem is a lot of people ask questions that claude et al can objectively answer way better than me, and they didn’t think to ask.
Should I a) give a worse answer b) ask claude and launder it as a meat proxy or c) tell them to ask claude?
b and c piss people off but often get them to a better answer.
In this scenario the answer for me is C. I say “Ask an agent about terms X, Y and Z and it should be able to explain this well for you. Try that and see if it works for you or if you have any follow up questions”
However, you have to distinguish when someone is asking for your judgement vs just general knowledge about a subject. If I’m asking for your judgement, you have to be the one delivering and owning that call in your own words. It’s fine if you verify or research your understanding with an LLM before answering, but I want to hear your voice and ownership not Claude’s. Whatever you get from the LLM you should understand and if possible, verify. If I came back and say “You said X, can you expand on that” you can’t reply “lol - Claude said that, I dunno”
How about being truthful? Say "I think A, but I'm not really an expert, so also try asking your favorite LLM"
What happened to "I don't know" or "I'm not sure"? Why do you need to direct someone to ask an LLM? If somebody asks me a question I don't know the answer to I'm not sending them anywhere, I tell them I don't know.
I got a spreadsheet of requirements for some software from a customer recently.
It was thousands of lines long with repeats and a boatload of conflicting requirements.
When we had a meeting to review it they didn’t know what many of the requirements were. They explained that many people made the spreadsheet but they were all in the meeting and nobody could quite describe some items or how they worked together. The vast majority of the spreadsheet was a mystery.
The phrasing and mishmash of concepts / inability of anyone to explain much of it made me suspect it was largely generated by AI.
- how come your blog page doesnt show a list of posts?
- https://gruhn.me/blog
- isnt that like standard behavior on any blog?
My pet peeve is endless Confluence Wiki pages that are clearly written by Claude (or whatever agent they like to use). I get bombarded with these, like it means something. Usually I say: "let me summarize this with my agent", but people rarely get the point.
I've been running across more and more of these. Many of my co-workers are falling for them. They'll run across one of these pages and start to get impressed by all the work a team has done and how organized they are. I dismiss it as slop and they defend it.
A week goes by and the topic comes back around. They actually took the time to read the page and realized everything was wrong and nothing made sense. They saw the page was updated (daily), went to go check the changes, and saw the whole page was rewritten each time, so they couldn't tell what changed. At this point they start to see the anti-value of all this, but keep falling for it every time they run across a new page.
I recently had experience working with one of the top security audit firms that collaborate with OpenAI on their cybersecurity tools. What struck me is that they found a lot of various issues (not all of them ended up to be legit, but a lot is) and filled the backlog of the project. However every collaboration with that
person was essentially a proxy for some security-related agentic prompt: - Hey, agent, open the issue! - Hey, I clarified it with the agent! - Agent disagrees! - Agent agrees!
Half of this feels like a communication/empathy problem. "Summarise this for someone without our context" fixes the wall of text, and just asking someone to do that might work.
Validation is harder, as nobody skips it because they weren't told to, but because relaying is faster, and unread output looks like validated output right up until it bites. That's probably not solved by asking nicely, I don't think there's a process fix for that beyond people actually being held to it.
Obviously the solution is to have a SKILL.md which will make an agent translate it to a humanese!
Lol, obviously never ventured into the Microsoft Techcommunity forums..
Actually I'm not sure why more people aren't reading and reposting outputs to save people tokens on generating more outputs (I gather that this article is more objecting to the reposting-without-reading)
I read this a couple weeks back and its appeal really stuck with me: if you expect someone to give their attention to something you made, make sure you've given your own attention and effort first.
https://news.ycombinator.com/item?id=48497609 https://tombedor.dev/human-attention-and-human-effort/
While I managed to avoid it for a long time, using generative AI to drive my day to day work life while resisting the pressure to become a meat proxy, I will admit that since joining an "AI native" startup I have become one. Whenever I'm trying to triage 20 different urgent things all day, every day, the economics of personalizing every message just isn't there anymore. It's not realistic. I do still try to way to filter down every AI-generated message into something digestible, but I feel like I'm in a mutually assured destruction type scenario where I can only survive the tidal wave of slop with more slop. I think we're in uncharted territory here in terms of the kind of workload generative AI can drown you in, especially when it's being used without strong discipline.
I do still think there are still places where being a meat proxy is legitimately okay or even desirable, such as PR comments. If everyone's code is generated by AI, and AI will ultimately be the primary consumer of the PR comments, it seems reasonable to me to allow AI to write the comments. Maybe I've just lost my mind though.
Expect to see more of this as people's writing skills atrophy. I have done this sometimes when the words aren't coming to me. But, yeah, it'd probably best not not bother with the comment if I can't come up with a coherent articulate comment myself.
Delegating stuff to Google (more recently AI) and repeating back the result, is one of my core functions at work next to programming.
A co-worker I'd just tell to RTFM when it's bothering me. Higher up the hierarchy, the egos are more fragile and I'm afraid the can't take it.
Rough opinion: If you're a developer who chose to be a meat proxy, you shouldn't whine if/when you're laid off, because clearly LLMs can do your job.
For anyone unfamiliar with the meat reference, here is Terry Bisson's delightful 1991 short-short story, "They're Made out of Meat":
https://www.eastoftheweb.com/short-stories/UBooks/TheyMade.s...
Some version of this needs to be hosted on a dedicated domain like nohello.com that I can respond with as needed. I might just do it
Last week a colleague did this to me; pasted to me in chat "Gemini said". I just ignored it and went with talking about the issue on another thread. I like LLMs and I use them every day, but I'm not gonna answer copy&pasted LLM stuff in human to human conversation. Distill it and let me know your thoughts; I don't need to read you LLM output, I can generate that myself.
I've been on the receiving end of too many "Slop Grenades" as well, to the point where I simply reply with https://noslopgrenade.com/ these days.
I have this set as my slack bio
I'm surprised this happens between workers within the same knowledge domain. If you're a software dev asking another software dev for an opinion on code, can't you just... talk shop?
In my experience this sort of thing happens all the time when something has to cross between different knowledge domains and the sender doesn't apply their mind in order to co-operate on the issue at hand.
In the pre-LLM past, the sender would add no value, acting as a simple email forwarder, and they would blame you for the delays caused by your inevitable clarificatory questioning. But I always had a defence, which is that your email had no inputs and I had to ask questions to clarify.
Now, though? With LLMs, they just run whatever it is (contract, memo, policy) through whatever LLM they have available and paste the output in an email, giving them the appearance of having done work and added value to the project.
But their LLM outputs don't make sense, or don't apply to our organisation, or is a fluffy and abstract "right answer" with no connection to the specific concerns of the business. Parsing it is a chore, and takes time.
And since now I am the only one actually taking that time, I become the visible cause of the delay.
It's infuriating. Any tips on how to deal with this would be greatly appreciated.
yeah similar thing happened to me
gov website changed, seems to have ditched buggy "new" data-collection features
said I couldn't find the menus even after logging in...
shares screenshot and 'expected menu paths' to get there, says "it's simple, look! claude says look there"
repeatedly asked "have you actually checked?" but... reply was "listen, I'm doing a big favor to you..."
...fking crazy...
some people are treating LLMs as some god-send truths
now, since this is coming from the C-level, I can't expect anything good to happen here
I use it to summarize my work and update JIRA/Asana because that is still the thing I hate doing most.
Of course you can relay the output. Dumb to say you can't. In the same way you can relay the output of anything, if it's useful.
But it should form part of your own judgment, not, as you say, just pasted into a chat without any thought.
That is super actual in my team. The biggest problem is that Claude tends to write long prose and use jargon which is not common for us. Often, to understand the whole idea I have to read all the text. The best if the writer rephrases what Claude has written, as wording/jargon will be familiar for the reader and will make communication easier.
I was talking to some friends that are heavy Claude users and suggested it was bullshitting, and got some pushback along the lines of it wasn't a person and didn't have agency. I'm not sure it's that simple.
The jargon can be functionally like bullshitting. It's preferentially aping language that's over everybody's heads in hopes of not being criticized. Whether that is a person's toxic habits (it's not, there's nobody in there) or an epiphenomenon of the design of the machine is less important: it's bullshitting if it's a social behavior there to manipulate people.
(jargon included where relevant, in hopes that people who know it will see it's the right word to use :) )
If you use the definition from On Bullshit, then bullshit is a statement made without regard for truth value because of its utility to the speaker.
As such, LLMs can only bullshit.
Is this the new "let me google that for you"?
> I can talk to Claude myself
OK but can you talk to my specific claude code conversation that has 6 months of context on the mechanism you are struggling with?
If you've been working on the issue for 6 months you should understand it well enough to answer in your own words.
I didn't say fix, I said context. As in "you are using an established thing the wrong way" which is most questions people ask. Edited
The biggest win for AI dev efficiency is cutting down what gets loaded into context. Semantically matching tasks to the top tools helps a lot.
> frequently contains all too plausible nonsense, and is increasingly jargon dense.
Does anyone know how to deal with the jargon part? It is getting hard to use claude and even worse when someone sends you the direct output from claude.
As with most things LLM, you can generally get what you want by asking for it (maybe not efficiently, but so it goes). "Once you have a answer, please summarize, and ask an agent to convert it to plain English" often helps.
I'm adding "ELI8" (explain like I'm 8 years old) to the end of my prompts a lot. It helps.
Why ELI8 and not the ubiquitous ELI5?
I will frequently post the output of a computer to slack as a part of an engineering discussion. For Claude output I'll treat it the same, wrap it in code blocks, and say something like "Claude's analysis". If its ok to post system logs, why is it less ok to post claude's output (also a computer), especially if I declare this? Is the difference that its plausibly human-level speech and thus breaks an implicit contract?
The problem is that there is no signal to the folks involved whether you have bothered to understand what you’re posting or are simply offloading that work to others.
Same issue as dumping a system log or a stack trace without any of your own analyses to be honest - people are going to wonder “okay. but what do you want me to do with this?”
But you're now just passing on the mental effort to decipher that analysis, without spending any effort of your own, onto someone else. Do you see the issue?
I append "sound kinda dumb but be factually correct" to my Claude requests and get much less jargony responses.
So you're telling me you haven't read the article?
I'll try that. I'm constantly telling it to be "brief, plain and on the point" or something similar. Tried caveman, but I think it had it's own issues.
But the jargon heavy very long responses are my main issue with it I guess.
Why are LLMs producing this super tense text more often now? Is it because they are being optimized to use fewer tokens?
I think they're overcorrecting for being too long-winded in previous generations (and still, in some cases). I guess this is a hard balance to get right.
ChatGPT is still plenty verbose by default IMX.
I'm pretty sure the other answers are wrong and it's a side effect of RL (see thinking machines post about inkling training). It's also exacerbated in fable and sol--I think it's token efficiency effect--bc it's about to reason with fewer tokens the density of the token information goes up.
I figure they are being optimized to write code/functions and not prose so all text is getting more code like.
I'll always give people my honest effort and benefit of the doubt initially, but those who violate it are treated likewise. The only way to put down this kind of behavior is to charge it a social cost. If we do not do this, the cost is externalized to everyone else who conducts themselves with care.
I happen to be currently working on our ISO 27001 implementation and as one of the prerequisites I'm writing our internal AI policy.
I've included a link to this blog post, just for kicks.
> "frequently contains all too plausible nonsense"
This really isn’t the case with frontier models in 2026.
I’ve found (sadly) that every time I thought the model was hallucinating, I was in fact the one who was mistaken.
Sure, your frontier model in 2026 will know in the team who is responsible for what part of the tech stack.
No?
Then, in the document where it writes who has to do what change, it is hallucinating. And this time you or me are not the ones who are mistaken.
N=1, and biased towards the type of questions asked. N+1, If I use one of the search engine sloptools I get frequent inaccurate answers which is I guess what the majority of people do.
I am prepared to admit that the model usually doesn't get things outright incorrect (unless you ask it to count letters). But the solution does contain a lot of nonsense. Not false nonsense, but meaningless or irrelevant sentences that make understanding the core of the fix much more difficult.
I am so tired of this. It’s even getting to the point where people without the full context and understanding are meat proxying incorrect information. Reading a bunch of slop that isn’t helpful hinders.
If they instead read it, and distill it down to “have you checked X?” Someone with the full understanding can easily go: “Yes, X doesn’t fit because of this other reason.” within seconds. Sometimes this happens without asking a teammate. However, when I do ask I’m not asking for you to ask Claude and paste the results, I have my own tokens for that. I’m asking you because I think your knowledge will be helpful in finding the answer.
Reminds me of this post, from a while back: https://news.ycombinator.com/item?id=48876441
claude said u cant call its users meat proxies. Now back to watching 'ow my balls!'
That so true. Every day I work with Claude Code I get a lot of words and acronyms and thought shortcuts I don't get. It's a flaw a big one
When it comes to cost per implemeting SPEC, agents are far cheaper than human salary. So, it's not going to go well.
That said thankfully there's no way of measuring a good SPEC yet. If there is that will collapse the SWEs profession.
Quite a few people do believe that a good specification is long, has multiple levels of headings, interspersed bold text, bullet points, and so on. Definitely, this is what they have seen happen throughout their lives: competent people have heaped praise on documentation which has had these signs, so this must be what good documentation looks like.
Nah, I disagree let them pay for my query dude. I want all the tokens I can get.
PS: Ideally set thinking to max and research mode or whatever anthropic calls it.
At least "Claude said" is honest. What's worse is when people send you slop and pass it off as their own work and expect you to take it seriously.
Claude told me to tell you, you’re right
Claude says this is funny.
Yeah, I just ignore people who do that. I don’t care what Claude or ChatGPT said. If you cannot be arsed to write yourself then I cannot be arsed to read it.
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
this is perfectly understandable?
This conversation is important but mostly they highlight transactional and most visible examples.
I can tell you even at one higher level of abstraction, this is happening a lot --
A business analyst supposed to ask discovery questions.
A project manager supposed to list risks and assumptions and a timeline.
A consultant expected to outline a new proposed solution and the pros and cons and trade offs.
An architect supposee to present and defend a solution design -- and the change review board supposed to ask questions to validate it.
All of them are lobbing slop grenades at each other -- just packaged in normal / human-looking formats like excel and word and powerpoint and email -- but slop nevertheless with minimal to zero value add by the human, thanks to proliferation of copilot (among others).
We are hastening the rot.
This is productivity theater.
This is where some of the 10x engineer and AI taking jobs is happening I am afraid.
Of course your grumpy coworker might then respond: "So if you can ask Claude, why did you ask me then?"
So the grumpy coworker understands that he is merely an unfriendly interface around Claude and suggests that he is to be made redundant accordingly?
No, the grumpy coworker was called upon for something you as his peer could have asked from Claude and not wasted his time.
Ridiciulous I know, but I've actually seen this happen.
soon enough most people will realize in that for software working in teams of 1-3 people is the best thing.
no reviewing 'a.i' code. talking to actual humans not 'a.i' jargons
This is mostly a problem with the product. It should write better. But maybe the priorities now are not on this part.
thank you for this post, this is absolutely happening i just did not have words to put on it and it is festering.
I agree. How can I see the rest of your blog?
"That's great for you, I'm not going to read it" is a phrase I have had to use.
This is the new "Let Me Google This For You".
We added this communication rule to our company handbook: "If you are asking for human attention, demonstrate human effort" (coming from https://news.ycombinator.com/item?id=48497609) and it's working well so far.
This is an interesting problem. A similar approach that worked for us was to only load the tools the agent actually needs for each task.
We're seeing this everywhere now, people generating thousands of lines of documentation or PR feedback in seconds, then dumping the actual labor of verification onto their teammates.
Thank you!
Finally I have a word I can use for this
I am sure this will lead to can you dumb this up prompt which will miss the important details
> Reading AI output is extra effort. It's verbose, frequently contains all too plausible nonsense, and is increasingly jargon dense. I recently got this sentence from Claude: "NATS control-plane events: stream leader election / R3 quorum re-form during pod churn." Jesus. I had to lookup almost every word to make sense of this.
I'm always puzzled when I see this kind of commentary. How are people using LLMs that they find this type of thing problematic? Just do:
- "It's verbose" -> prompt "Explain briefly"
- "plausible nonsense" -> Yes, it happens, but less frequently than with older models, and arguably far less than your average human. Mitigate by prompting "Run a search to check that X, Y, Z is correct".
- "I had to lookup almost every word to make sense of this." -> Does this really happen frequently? It doesn't to me. Mitigate by prompting "Breakdown that sentence and explain each term in plain English".
Too many damn people are meat proxy
Weve been writing articles like this since 2023 and the problem only gets worse year-after-year.
I’m sorry AI purists but you have lost the argument. I know a guy who just started at my firm and he uses LLMs all day to submit his work product.
He’s getting ahead. Coworkers don’t seem to mind it.
It is time to come to grips with the fact that people are going to be “meat proxies” and be promoted over you unfortunately. Whining on HN hasn’t changed this dynamic and it never will.
Claude's not a person. Claude is just a tool.
It's my very same exact opinion and as an EM I'm trying to push this everywhere (team, peers, upwards) but it feels like speaking to a rubber wall. It starts to be exhausting and infuriating. This is not an "LLMs are bad" issue, this is a "humans are lazy" issue. (shit, I wrote the it's not X it's Y construction!)
Friends don't make friends read slop.
If I have something Claude wrote then I will offer the LLM output. But I generally prefer to at least give the option.
I have seen conversations where it has boiled down to an exchange of A's Claude responding to B's Claude with a retort and fixes from A's.
I've gone another route when I found myself simply replying with an investigation made from an AI: Actively encouraged my coworkers to ask the knowledge-bank AI I've setup instead of asking me. It has read-only credentials for all our environments and access to our non sensitive codebases, so it can answer any question much faster and accurately than I can. It works too! We use grafana less, we get more accurate bug reports, we understand the bugs/issues far better.
The "problem" here is that AI is too great at providing answers, so lean into it instead of fighting it. It won't go away.
eventually two people will just be forwarding AI outputs to each other, each assuming the other one read it
In these situations I always use AI to respond
The question is what is reasonable. This is more how I work for code and I just tried for a document.
i) Prompt (Human input)
ii) Response (LLM)
iii) For (i=0;i!=n;++i) {
iv) Review/adjust (Human input)
v) Response (LLM(s) - different perspectives) }
vi) Rewrite parts that failed/I am not happy with (Human input).
vii) Optionally repeat from iii.
This tends to get me somewhere I want. For code at least. I made a document with it recently (fairly concise) and still worried I'd missed some AI slop in it. Though there is always human slop if I just made it by myself.
The problem is that people either don't understand how economic value is created, or want to obscure their lack of contribution to it. They see that they can make the AI do some work without revealing what work they put in. They use the AI as a shield. To genuinely and clearly provide value they have to reveal their prompt and verify its response.
It's our absolute obligation to make sure "meat proxy" sticks as hard as "vibe coding" did. It's too good not to.
That is problematic, even if the intention is to be ironic and to critize a recently witnessed practice, this only will make leave everything human out of the conversation
When using AI to generate content, content of any kind - then I urge the author to evaluate the true cost for the organisation, not just for him or her. Using AI to generate 20 pages of documentation in 30 minutes looks good for you - but having 100s of people read through that slop will cost the organisation, not only in plain work hours, but also in lost trust, respect, joy of work as well a sense of purpose.
And yes, I've gotten '50 pages tech spec' docs sent my way from upper management -- as a 'help'....
When "ai slop" became a concept, it mostly came into spaces where "human slop" had already been firmly established.
Obvious ones like content scroll fodder. But also, there's slop everywhere. Even things like longform journalism or trade publications... these are often "slop." A string of platitudes and generalizations that superficially seem serious.
That is true in work culture too. A lot of people jump in on emails, superficially seem to be working... but contribute little of value. Now they do this, but with ai.
The copying and pasting of AI output into a chat has previously been referred to as a “slop grenade”. That’s a good phrase for it!
Remove meat proxy before flight
> I can talk to Claude myself. It's going to be faster and I get to control the context. I don't need a meat proxy in between.
Yes you do apparently need a meat proxy, because either 1) you didn't tell them that you already talked to claude, or 2) you didn't already talk to claude. Either way they are forced to be your meat proxy because, apparently, you never asked an AI, or didn't tell them you asked an AI, and asking an AI is now table stakes.
"The CI is failing on the main branch"
Don't be a meat proxy!
Lets not be stupid about this. Do whatever works for you, and don't let some moron set the standards for your social interactions because they got the ick one morning, just because they wrote it down and put it on a website.
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
> Jesus. I had to lookup almost every word to make sense of this.
There's a lot to say about this, but I would just start with why are you using NATS if you don't understand it at all, and how did this become AI's fault?
well, I dont understand code myself, so in this case the only way I could to do verify is to test the real product
I wrote about this a year ago: https://blog.gpkb.org/posts/just-send-me-the-prompt/
Sending LLM outputs to people is worse than useless, because you don't even know what they prompted with. For all you know it was "write me a paragraph justifying my opinion".
Or you could learn to verify what they sent you. Geezus fuck. How about knowing why did they ask this, and not some other question.
When you say "X proxy" the 'X' stands for the thing that is being proxied not the thing that is doing the proxying.
You either get the AI notice or you don’t. The AI output will still be posted.
It’s a meat to meat courtesy during this transition phase.
I think meat proxies are solved by ARs which is the new PR, we built a tongue-in-cheek attentionrequest.com site about it, but the general concept has been working for our team -- asking people to write questions before they send an agent-generated doc, to serve as a proxy for having read / thought about it.
This is a weak argument but I hope you develop it into a better one.
Genuinely open to that. What's weak about it?
There is no meat proxy subverting discourse. There are just a bunch of dumb disinterested people in the way of a real community.
The real reason people do it:
1. It's quick
2. It's not their problem anymore because they've replied.
No amount of hand-wringing is going to change that.
And nobody gives a fuck at work anymore.
Directives want results, positive numbers and growth. How well done is and the queality of the product is something trivial.
Until february/april I was super dead serious about the quality of my apps and my code, I was super strict and always looking for excellence and quality. But people were delivering shit faster than me and someone who doesnt give a fuck about anything of this just got promoted. So now i just let opencode do my work while I watch and play dota in the background. I deliver faster and nobody has ever talked about quality of my code or the ellegance of it.
I still programming by hand in my personal projects, but in corporative projects? Fuck it, i'll let the llm do my job until it lasts and Ill watch the world burn.
Based
Recent models have had a worrying lean towards longer output. If you're going to serve up some AI slop, at least distil it to the smallest possible footprint possible.
in the same vein, I have a coworker who's code reviews have magically transformed into stopping-the-world with all these great ideas for abstraction and insane edge cases... he must have gotten really good overnight!
but seriously, when I get a review like that, my first thought is that why would I read a summarized AI output, which I just did for 5 hours while whipping Claude to stop re-writing raft from scratch.
If people say: AI said "X". It usually means you should've asked AI before going to the human.
A lot of odd takes on this, the only times I have said “Claude said: …” was equivalent to “let me google that for you” in the olden times. If someone asks me something and by pasting their question into Claude I get the correct answer it means that person needs to learn how to use the tools available to them better.
Well put. After reading, I came here to say the same thing.
The times that I "Claude said:" are almost always a case where Claude gave the correct answer.
The real problem is if you paste "Claude said:" without reading what Claude said. Have seen this happen and goes something like:
Sr Engineer: "Claude said:"
Me: "But that doesn't make sense because of X"
Sr Engineer: (having actually engaged thinking effort: human) "Oh yeah, not sure why Claude said that, I think we should do Y instead"
I do this often for stupid questions. It's like sending someone a lmgtfy link.
Don't waste my time by asking me a question that:
A.) I know no more about than Claude, and
B.) You could have answered yourself using Claude
All of the questions I respond to this way fulfill both A and B. Responding with "Claude said" is both more polite than most ways to respond to such questions, and a subtle hint to the other person that they should be using the LLM first for those types of questions, rather than coming to me first.
So, don't ask you anything? Seems I'd rather have your opinion if you know more than Claude.
>So, don't ask you anything?
No. Don't ask me stupid questions that Claude knows more about than me, and that you could've answered yourself with Claude.
>Seems I'd rather have your opinion if you know more than Claude.
I get a lot of questions about subjects I know considerably less about than Claude, and the subset of those which can be answered by Claude (which isn't all of them, to be clear) are the questions I respond to this way.
Just my boss..
tbf rephrasing the output in my own words has become the best and most important tool in ensuring quality from AI. Whether it's an answer to a question or an implementation of some ACs that I copypasta'd in, the best sanity check is understanding and the best demonstration of understanding is the ability to redo.
Yes
im the opposite: be a good meat proxy, not a bad one.
I call it BI Slop now [0]. It's everywhere. Usually "claude said" is used a a proxy for authority, as in there is some validity in this statement. But it's disguised as humbleness. i.e. I don't know if this is true, but claude said xyz, with your expertise you might know more than me...
[0]: https://idiallo.com/blog/business-intelligence-slop
niceee
The policy I set at my org is to put an organically written message at the top of your Claude blob. Something like:
tl;dr: ___human-written____
____claude blob here_____
Where the idea is that humans only read human writing, and Claude is able to share context between sessions with the jargon-riddled blobs.
Trouble is, I catch people every day blatantly writing the Tl;dr with Claude or ChatGPT.
People do this all the time though w "paul graham said" "elon musk said". Why not "Sum Total Avatar of All Human Knowledge Said ..."
[flagged]
[flagged]
[dead]
[dead]
[dead]
[flagged]
[flagged]
Don't spread ignorance like this.