If you really rewrite LLM’s response in your own words, I will know that you have learnt something.
Because if you tell me directly that you have asked Claude, next time I will probably ask Claude directly as I don’t need you.
And we won’t be able to distinguish what is yours and what is claude’s so I’ll be subconsciously suspicious that the whole answer is ai-generated (/skill me-persona-answer-descriptive)
That is the reason why doctors wear white and have stethoscope. In many cases people don’t argue with their opinion as they know that doctor had to spend 6 years to earn it. But if they admit LLM as a source they are becoming replaceable.
The emphasis should be on “rewriting”, even kids know copy-paste and it doesn’t count :)
and what if i tell you i asked stack overflow?
What we’re seeing now is industrial grade ignorance that can only be observed in in-person or video meetings.
Asking AI sometimes gives you the same answer as AI is trained on these same forums, but not always.
Your prompt structure and/or inference bugs (which is a lot more common in smaller providers or local hosting) can change the answer AI gives.
And ofcourse, if there's low/no data, AI will still give an answer even though it's not in the safe zone.
We don’t usually tolerate copy-paste answers at school so why should it count at work?
I've found that it's a thing that develops over time with experience, though. And I want people I'm training to develop that skill rather than farm it out to an LLM that never gets it right.
What is this macho bullshit of pretending like you have memorized all information you might ever need and looking something up is a sign of inadequacy?
And yes Claude or whatever is just another source, to be verified just like any other.
Or, if you get nerd-sniped by the question and spend some time figuring it out, that’s fine too.
But if you want to be helpful but don’t want to take the time to figure it out yourself, don’t just forward the question to AI or send me a link to the first result in Google because I could have done that myself(and may have done it already). Just say you don’t know, which is a paradoxically more useful response.
It shows someone where they can find that information for themselves in the future. That way they don't have to bug you later if they forget and it can give them a useful resource they can explore. If nothing else it demonstrates that at least one other person had the same understanding of something that I did which could be reassuring.
On the other hand, it's nice when someone tells you an answer is AI generated so that you can apply an appropriate level of skepticism to that answer. Maybe you can even reply to let the person know when inevitably the something they just "learned" was entirely bullshit.
Part of the problem with people sending text/screenshots right out of AI chatbots is that it suggests that not only were they so lazy that they went to a pathological liar chatbot instead of thinking about what you asked, but they likely didn't bother to review/fact check any of it
What a waste of time. Do you treat your coworkers like students trying to do homework that you assigned?
> Because if you tell me directly that you have asked Claude, next time I will probably ask Claude directly as I don’t need you.
I should be upfront about the use of AI when providing an answer. You in turn would learn that you can find the solution using AI yourself next time. You learn something and so do I. What is wrong with this interaction?
> That is the reason why doctors wear white and have stethoscope. In many cases people don’t argue with their opinion as they know that doctor had to spend 6 years to earn it. But if they admit LLM as a source they are becoming replaceable.
And don’t get me started with doctors. I hate when someone just tells me to trust them just because they are a professional. If they are worth their salt then they can defend their position without using their costume or degree to intimidate me.
I’m not condoning simply taking the first answer that AI spits back at you and regurgitating that as a response. But if the answer is correct then there’s no need to rewrite it. I just feel that rewriting the answer is trying to hide the fact that you had to use AI to help you find the solution, which to me, is dishonest.
The fact that people vet the AI answer before responding is the value added not the process of rewriting the response to protect frail egos.
They seem to think they've more or less solved the problem by posting an LLM's response to the issue or concern I've raised.
Lots of people aren't very thoughtful or wise, including some supposedly very intelligent people.
For further proof: think of all the workers proudly parroting their bosses' anti-union rhetoric, like they're temporarily embarrassed billionaires.
> The revised bill arrived: $1.00 for turning the screw; $9,999.00 for knowing which screw to turn.
In many a company, the "old neckbeards" and "dead weight" are the first ones to be cut or eventually be driven off by ever more outright bullshit - and often enough, it is only realized way too late that important "institutional knowledge" is gone [2].
[1] https://calvincorreli.com/blog/1397-knowing-which-screw-to-t...
The right way to solve this is for the PM to forward it to the ME who designed the brobillator and calculated the 1-month maintenance interval, CCing the controls engineer who helped ensure that machine wouldn't eat itself with the fault and low-speed mode.
The wrong way to solve this is for the PM to forward it to ChatGTP[sic], which might mindlessly suggests maintenance-free sealed bearings (that are totally inadequate for the temperatures and contaminants). If he asks the mechanical engineer to redesign it like ChatGTP suggested, that situation may be salvageable. If he told the customer that they'd have an engineer out in 3 days with new sealed bearings it's bad.
“I don’t really know much about that, go ask _____” is the desired response in that situation
AI answers may or may not be completely hallucinated, and often the people copy/pasting them didn't even read them.
Back when most Google results were authentic web-pages, something like "here's a web page that I think solves your problem" was a fairly useful reply from a coworker.
At least with the example in the article (with the ChatGPT screenshots), I don't think it's all that different from the olden days when people would include links to an unnvetted webpage after a quick web search, or a link to something like let me Google that for you. It isn't about feeling like they contributed. It's more a passive aggressive way of saying do your own research.
"Interesting question, I asked Claude that question, and here's what I got for a response. Here's what I thought was interesting about Claude's response and what I think applies. What do you think?