Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Large language models reduce public knowledge sharing on online Q&A platforms

academic.oup.com

Post reply on HN
Print view View original

181–190 of 366 posts

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#181

Post by jancsika »

haolez wrote: post #72
PhilipRoman wrote: post #55

Earlier quoted context omitted.

I haven't really found stackoverflow to be that humiliating (compared to some IRC rooms or forums), basic questions get asked and answered all the time. But the worst part is when you want to do something off the beaten path. Q: how do I do thing X in C? A: Why do you need to know this? The C standard doesn't say anything about X. The answer will depend on your compiler and platform. Are you sure you want to do X ins…

The first time that I asked a question on #cpp @Freenode was a unique experience for my younger self. My message contained greetings and the question in the same message. I was banned immediately and the response from the mods was: - do not greet; we don't have time for that bullshit - do not use natural language questions; submit a test case and we will understand what you mean through your code - do not abbreviate…

This being HN, I'd love to hear from one of the many IRC channel mods who literally typed (I'd guess copy/pasted) this kind of text into their chat room topics and auto-responders.

If you're out there-- how does it feel to know that what you meant as a efficient course-correction for newcomers was instead a social shaming that cut so deep that the message you wrote is still burned verbatim into their memory after all these years?

To be clear, I'm taking OP's experience as a common case of IRC newbies at that time on many channels. I certainly experienced something like it (though I can't remember the exact text), and I've read many others post on HN about the same behavior from the IRC days.

Edit: clarifications

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#182

Post by phil-martin »

blazing234 wrote: post #111
zmgsabst wrote: post #100

Earlier quoted context omitted.

You can usually even ask the same LLM: - do a task - criticize your job on that task - redo that task based on criticism I find giving the LLM a process greatly improves the results.

How do you know the second result is correct? Or the third? Or the fourth?

I approach it the same way as the things I build myself - testing and measuring.

Although if I’m truly honest with myself, even after many years of developing, the true cycle of me writing code is: over confidence, then shock it didn’t work 100% the first time, wondering if there is a bug in the compiler, and then reality setting in that of course the compiler is fine and I just made my 15th off-by-one error of the day :)

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#183

Post by iSnow »

zmgsabst wrote: post #100
delichon wrote: post #85

Earlier quoted context omitted.

I've gotten answers from OpenAI that were technically correct but quite horrible in the longer term. I've gotten the same kinds of answers on Stack Overflow, but there other people are eager to add the necessary feedback. I got the same feedback from an LLM but only because in that case I knew enough to ask for it. Maybe we can get this multi-headed advantage back from LLMs by applying a team of divergent AIs to the…

You can usually even ask the same LLM: - do a task - criticize your job on that task - redo that task based on criticism I find giving the LLM a process greatly improves the results.

That's a smart idea I didn't think of.

I've been arguing with Copilot back and forth where it gave me a half-working solution that seemed overly complicated but since I was new to the tech used, I couldn't say what exactly was wrong. After a couple of hours, I googled the background and trust my instinct and was able to simplify the code.

At that situation, where I iteratively improved the solution by telling Copilot things seem to complicated and this or that isn't working. That led the LLM to actually come back with better ideas. I kept asking myself why something like you propose isn't baked into the system.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#184

Post by yieldcrv »

herval wrote: post #169
yieldcrv wrote: post #162

Earlier quoted context omitted.

The models come out fast enough Doesn’t seem to be a great strategy to always need these things retrained, but OpenAI’s o1 has things from early 2024 Don’t ask about knowledge cutoffs anymore, that’s not how these things are trained these days. They don’t know their names or the date.

Not my daily experience. It’s been impossible to get relevant answers to questions on multiple languages and frameworks, no matter the model. O1 frequently generates code using deprecated libraries (and is unable to fix it with iteration). Not to mention there will be no data for the model to learn the new stuff anyway, since places like SO will get zero responses with the new stuff for the model to crawl

Yes I encounter that too but for things in just the last few months with o1

It is really difficult if you need project flags and configurations to make things work, instead of just code

Github issues gets crawled, where many of these frameworks have their community

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#185

Post by jneagu »

melenaboija wrote: post #38

It’s been a relief to find a platform where I can ask questions without the fear of being humiliated Half joking, but I am pretty tired of SO pedantry.

I am very curious to see how this is going to impact STEM education. Such a big part of an engineer's education happens informally by asking peers, teachers, and strangers questions. Different groups are more or less likely to do that consistently (e.g. https://journals.asm.org/doi/10.1128/jmbe.00100-21), and it can impact their progress. I've learned most from publicly asking "dumb" questions.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#186

Post by ElFitz »

PhilipRoman wrote: post #55
melenaboija wrote: post #38

It’s been a relief to find a platform where I can ask questions without the fear of being humiliated Half joking, but I am pretty tired of SO pedantry.

I haven't really found stackoverflow to be that humiliating (compared to some IRC rooms or forums), basic questions get asked and answered all the time. But the worst part is when you want to do something off the beaten path. Q: how do I do thing X in C? A: Why do you need to know this? The C standard doesn't say anything about X. The answer will depend on your compiler and platform. Are you sure you want to do X ins…

It’s also quite fun when you ask niche questions that haven’t been asked or answered yet ("How do I do X with Y?"), and just get downvoted for some reason.

That’s when I stopped investing any effort into that community.

Turned out it, counter-intuitively, was impossible. And not documented anywhere.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#187

Post by ocular-rockular »

jneagu wrote: post #185
melenaboija wrote: post #38

It’s been a relief to find a platform where I can ask questions without the fear of being humiliated Half joking, but I am pretty tired of SO pedantry.

I am very curious to see how this is going to impact STEM education. Such a big part of an engineer's education happens informally by asking peers, teachers, and strangers questions. Different groups are more or less likely to do that consistently (e.g. https://journals.asm.org/doi/10.1128/jmbe.00100-21 ), and it can impact their progress. I've learned most from publicly asking "dumb" questions.

It won't. If you look at advanced engineering/mathematics material online it is abysmal in quality of actually "explaining" the content. Most of the learning and understanding of intricacies happens via dialogue with professors/mentors/colleagues/etc.

That said, when that is not available, LLMs do an excellent job or rubber ducky-ing complicated topics.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#188

Post by beachy »

Aurornis wrote: post #161
melenaboija wrote: post #38

It’s been a relief to find a platform where I can ask questions without the fear of being humiliated Half joking, but I am pretty tired of SO pedantry.

Many of the forums I enjoyed in the past have become heavily burdened by rules, processes, and expectations. They are frequented by people who spend hours every day reading everything and calling out any misstep. Some of them are so overburdened that navigating all of the rules and expectations becomes a skill in itself. A single innocent misstep turns simple questions into lectures about how you’ve violated the rule…

> Some people appear to thrive in this environment where there are rules for everything. People who memorize and enforce all of the rules on others get to operate a tiny little power trip while opening an opportunity to lecture internet strangers all day.

Toddlers go through this sometimes around ages 2 or 3. They discover the "rules" for the first time and delight in brandishing them.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#189

Post by VancouverMan »

MASNeo wrote: post #39

Wondering about wider implications. If technical interactions reduce online, how about RL and how do we rate a human competence against an AI once society gets a habit from asking an AI first? Will we start to constantly question human advice or responses and what does that do to the human condition. I am active in a few specialized fields and already I have to defined my advice against poorly crafted prompt response…

> Will we start to constantly question human advice or responses and what does that do to the human condition.

I'm surprised when people don't already engage in questioning like that.

I've had to be doing it for decades at this point.

Much of the worst advice and information I've ever received has come from expensive human so-called "professionals" and "experts" like doctors, accountants, lawyers, financial advisors, professors, journalists, mechanics, and so on.

I now assume that anything such "experts" tell me is wrong, and too often that ends up being true.

Sourcing information and advice from a larger pool of online knowledge, even if the sources may be deemed "amateur" or "hobbyist" or "unreliable", has generally provided me with far better results and outcomes.

If an LLM is built upon a wide base of source information, I'm inclined to trust what it generates more than what a single human "professional" or "expert" says.

Re: Large language models reduce public knowledge sharing on online Q&A platforms

#190

Post by bayindirh »

PhilipRoman wrote: post #55
melenaboija wrote: post #38

It’s been a relief to find a platform where I can ask questions without the fear of being humiliated Half joking, but I am pretty tired of SO pedantry.

I haven't really found stackoverflow to be that humiliating (compared to some IRC rooms or forums), basic questions get asked and answered all the time. But the worst part is when you want to do something off the beaten path. Q: how do I do thing X in C? A: Why do you need to know this? The C standard doesn't say anything about X. The answer will depend on your compiler and platform. Are you sure you want to do X ins…

Yes, immature people are everywhere, but SO took it to a new level before they had to implement a code of conduct. I remember asking questions and getting "this is is common misconception, maybe you're looking for X instead" type of actually helpful and kind answers.

After some point it came to a point that if you're not asking a complete problem which can't be modeled as a logic statement, you're labeled as stupid for not knowing better. The thing is, if I knew better or already found the answer, I'd not be asking to SO in the first place.

After a couple of incidents, I left the place for the better. I can do my own research, and share my knowledge elsewhere.

Now they're training their and others models with that corpus, I'll never add a single dot to their dataset.

Post reply on HN