Comment by deepsummer
15 hours ago
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
> Is it not more akin to asking a handyman to go and do the work for you?
Does your handyman go off the rails and build pipes around the neighbors house because they didn’t know you already have an incoming water pipe?
Does your handyman require you to have expertise in plumbing to make sure they don’t use bubble gum to seal the pipes?
It’s not taking credit if the “handyman” is a gibbering idiot who needs more handholding than a child.
1 reply →
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.
Even Mythos can hallucinate[1] with a codebase to analyze. The argument I often see is that humans are fallible, too, but the issues we're seeing in this topic are about those who should know better—including in senior positions—heeding LLM responses/advice over human and concerningly not actually thinking about things at all. Ie: they're not being treated by many as just useful tools/tentative feedback but as authoritative answers/solutions.
There still needs to be critical thinking involved on the human side, even if there's a high rate of accuracy in certain dimensions.
[1] https://news.ycombinator.com/item?id=48434824
1 reply →
That just isn't true. LLMs still hallucinate all the time.
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: "
Maybe not just "Claude said". If one of the issues is getting the right prompt, you might say what prompt you used.
Sometimes I do that. The problem is that it makes them look bad.
2 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)?
Literally just ignore these people if they don't get the message. Let them complain to their supervisor and hopefully their supervisor will bring up the fact that you asked them to use AI instead. It's an easily defensible position. Some people need to learn the hard way.
Public embarrassment is also good. I've done it and had apologies off people. They tend to get the message pretty quickly if they're made to look dumb as fuck.
2 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.
Yesterday, when I was working with LVGL, using Fable.