Comment by bayindirh
10 hours ago
Many of us understood that, and tried to tell it over and over. My keyboard is worn, my mind is numb, my soul is weary from typing/telling that the output is not the aim, but the content it carries, the substance it holds.
Yet a very vocal group advocates integrating LLMs into every aspect of their lives, and the other group watch in dismay.
On the other hand, I used a sentence form used by LLMs in a comment during a morning-brain moment, and an angry mob tried to crucify me with torches and pitchforks.
Many people love to generate text with LLMs, but most of same crowd apparently hates to read LLM output. An interesting phenomenon.
It must be a distinction between engineer-type brains and leader-type brains. If you're focused on social status climbing, the fact that you produced a report, and the length of that report, is far more important than whether the contents of the report have any relation to reality beyond the surface level. This is true if you're trying to get VC funding as well. It's even in institutional academia, as seen by the Sokal affair.
> is far more important than whether the contents of the report have any relation to reality beyond the surface level.
This only works in broken companies where leadership is already overrun with other status climbers.
If someone tries this trick at a functional company and is discovered, their reputation takes a huge hit that can take years to overcome, if ever.
At high performing companies reputation is everything. It can be a real problem to even get people to work on difficult or risky problems because they’re afraid of having a failure attached to their reputation.
Throwing unreviewed LLM slop around would be a reputation death sentence if your peers are good at what they do, because by now we all know that it’s easy to generate long reports with LLMs. The secret is out and anyone who follows technology knows it.
There are some companies where you can LLM report-slop your way to prestige. Those are not companies you want to stay at.
> This only works in broken companies
The promotion of LLMs is, I believe, basically a referendum on how many companies are broken in that way. So far, the adoption statistics are looking like it’s most of them.
I thought for a second you were going the opposite way with the brain thing. Good leaders are very much concerned with accurate and timely communication.
It can be a very engineer thing to point at a vanilla Doxygen dump and call that thing "documentation".
It is documentation. This week, I looked at the Laravel API docs[0], just as much as I looked at the framework guide[1]. It all depends on the question you have and the answer you need.
It's like the car manual (Do they still ship with one?) where every detail is laid out. It's not important for daily driving, but when you to fix an electrical problem, that's the reference you need.
[0]: https://api.laravel.com/docs/13.x/index.html
[2]: https://laravel.com/docs/13.x
1 reply →
To be frank, I neither live/work in an environment or live for "climbing" anything, so fluffware is not something I produce or understand.
On the other hand, I have friends who play that game, and I don't talk with most of them, because their life views and mental states continually drift to a state which makes no sense whatsoever for me, so we grow apart.
Eventually, enough engineers will breed with enough engineers, that we'll be a different species, incapable of communicating with weirdo humanity.
3 replies →
> but the content it carries, the substance it holds.
It's even more than that, I think. It's the intent to reach out and transfer to others something that you experience internally. It's a sense of purpose motivating that content and substance to be packaged up and communicated in the first place.
Sometimes LLMs will slip up and say things like, "I see this working with people all the time." And it REALLY breaks the illusion that the text they extrude means anything. Because obviously there is no "I" and no, the LLM does not have this autobiographical memory the text is simulating.
My wife uses ChatGPT a lot for advice on child development and parenting in general. Every so often, it says things like "in my experience, what I've noticed works best is..."
> Many people love to generate text with LLMs, but most of same crowd apparently hates to read LLM output. An interesting phenomenon.
Are you sure it's the same people? I feel the opposite: it's the same people who happily read tons of LLM output themselves and send some of it to others.
my coworker presented a 93 page pdf for a project plan entirely created by $230 of Claude's deep learning. i hope i wasnt too mean while asking him variations of "what now? what do i do with this?" the answer is obvious: read it. but so much of it covers problems which are already solved and my coworker clearly hasnt considered at all
point is: maybe llm output is best not shared. maybe, like a dream, the only audience who can appreciate the content of the writing is the first party
Having made the mistake of writing a couple of them, I’m pretty sure the only uses of non-fiction documents longer than ~30 pages is
* As a reference; people can look look up just the algorithm they want
* As a token that says “I’ve thought about this problem”
(The former is somewhat a subset of the latter. All of the stuff they didn’t read when they used the work as reference was there to provide them confidence that the algorithm they plucked out of the thing had some thought behind it).
totally. ive made some 10 pagers documenting topics ive THOROUGHLY considered and had my work peers review. same as you, the purpose is to point at it 5 years from now and say "the answers in there"
but when my coworker pays to create documents on topics he hasnt considered, for ideas of projects which dont exist, it feels like an exercise in text generation and i become mildy put off for the onus of reading it. i know it was written without purpose and while there are possibly novel tech and ideas in there, it feels like being handed less than a rough draft because it lacks any aim and intention
now, i expect he would get much more from it. hes the first party audience who reads the doc and sees a valuable project somewhere inside of it
Sometimes the medium is the message though. Especially in literature, prose matters. It's not just about what is said but also how.
This applies even in "low" art. AI generated imagery from mainstream models tends to be so ugly it detracts from effectively communicating what the prompter intended.
I don't we disagree a little bit, I think. Content and substance also contains "how" alongside "what". At least that was my intention.
I stopped sharing my photos publicly to stop supplying training data for image generation models. I don't want to feed that machine.
Most people love to drive, but hate to be stuck in traffic. Or for the countryside to be all freeways and the city to be all parking lots.
It's not the exact same asymmetry, but there are parallels.
No one likes generating content. Many like generating income or status. It is a new era. The era of curation. I would predict shorter and weirder stories and meme style communication patterns outside normal syntax. Poetry e.g. is replicable but not redoable. I am ambitious we are to see the best writing ever soon.
> No one likes generating content.
Oh no. I see the contrary.
One of my friends recently stated (paraphrased):
And they continued how they replace almost everything they do with LLMs.
I have seen people who never think but spend all their days querying LLMs for everything. Docs, help, howtos, system administration, coding...
What I see is an incentive for less work, less effort. Not income or status. They don't spend their "saved" time, if there's any, for personal activities or self-improvement either. They just do things with less effort, and just live as they lived before.
It's brain numbing.
> I would predict shorter and weirder stories and meme style communication patterns outside normal syntax.
I don't share the same sentiment, yet I don't have idea what to expect. It has to get darker to get brighter again.
We're just starting.
It is very weird. 6 months ago I generated some documentation for a service that was largely undocumented. When it was done my only thought was "who is this for?" Is any one going to actually read this or are they just going to ask the LLM? If someone didnt write it, why would I read it? It just becomes something to pipe into the context for the next person. It seems though a lot of people don't have this experience, its like they think they are the only ones with access to this technology.
Additionally I am so, so ,so sick of reading claudes overly wordy technical writing ish bullshit. Its exhausting trying to find the fuck ups or hidden assumptions in the pages of text it produces for everything.
Every time I try and search for ways to make it stop I just find people trying to save tokens which I don't give a shit about. I just want it to communicate things to me without having to constantly read 4 paragraphs, dump the context or fork it or start a new session so it "remembers" or re-weights that is was told to be concise and terse in like 7 different places. Maybe I am using it wrong.
In the rare occasion I use LLMs, I always add at least one of the following:
Then I skim the result, and jump to the references it has given. I don't have time for other way.
You have reinvented the search engine.
That’s not meant to be glib; Google and friends are pretty trash now.
2 replies →
To be fair, LLMs didn't really change this that much: Throughout my career, I've been generating docs that nobody is going to read, only because some process somewhere says "In order to get engineering approval to commence work, an Engineering Risks Document (ERD) must be produced." Everyone knows that nobody reads these, but someone with a rubber stamp checks to see that the doc exists and it looks sufficiently long, and then he stamps the stamp. LLMs just make producing this shit faster.
For my medical notes assistant I had to set a global "write in a telegraphic, minimal style" because it was fluffing up every aspect of the notes. Nobody wants to read that BS and it's impossible to find information in a series of walls of text.
It can also go the other way. It outputs a paragraph so full of telegraphic technobabble that it’s just unreadable, inventing new jargon every other sentence, using terminology invented in its thinking stream and never explained. And when it combines the two issues and it outputs 4 pages of technobabble, it gets just exhausting to read. Paradoxically, given that 2 years ago everyone was using LLMs to summarize web pages, their ability to condense text (including their own thought stream) is still lackluster. I dream of some post-processing diffusion optimizer just condensing all of their slop away. Until then, I just tell it to _avoid_ telegraphic and newspaper-title speech and to output one or two paragraphs of prose only.
I wonder if reading cooking recipe sites in training contributed to this.
9 pages about how it felt to be a kid eating the dish, for 7 ingredients and a one line recipe.