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

151–160 of 366 posts

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

#151

Post by elicksaur »

epgui wrote: post #137
insane_dreamer wrote: post #120

The problem is eventually what are LLMs going’s to draw from? They’re not creating new information, just regurgitating and combining existing info. That’s why they perform so poorly on code for which there aren’t many many publicly available samples, SO/reddit answers etc.

In a very real sense, that’s also how human brains work.

This argument always conflates simple processes with complex ones. Humans can work with abstract concepts at a level LLMs currently can’t and don’t seem likely capable of. “True” and “False” are the best examples.

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

#152

Post by jneagu »

antisthenes wrote: post #138
mycall wrote: post #125

Earlier quoted context omitted.

I thought synthetic data is what is partially training the new multimodal large models, i.e. AlphaGeometry, o1, etc.

Synthetic data without some kind of external validation is garbage. E.g. you can't just synthetically generate code, something or someone needs to run it and see if it performs the functions you actually asked of it. You need to feed the LLM output into some kind of formal verification system, and only then add it back to the synthetic training dataset. Here, for example - dumb recursive training causes model collaps…

Anecdotally, synthetic data can get good if the generation involves a nugget of human labels/feedback that gets scaled up w/ a generative process.

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

#153

Post by roughly »

stickfigure wrote: post #133
insane_dreamer wrote: post #120

The problem is eventually what are LLMs going’s to draw from? They’re not creating new information, just regurgitating and combining existing info. That’s why they perform so poorly on code for which there aren’t many many publicly available samples, SO/reddit answers etc.

> The problem is eventually what are LLMs going’s to draw from? Published documentation. I'm going to make up a number but I'll defend it: 90% of the information content of stackoverflow is regurgitated from some manual somewhere. The problem is that the specific information you're looking for in the relevant documentation is often hard to find, and even when found is often hard to read. LLMs are fantastic at reading…

Yeah, this is wildly optimistic.

From personal experience, I'm skeptical of the quantity and especially quality of published documentation available, the completeness of that documentation, the degree to which it both recognizes and covers all the relevant edge cases, etc. Even Apple, which used to be quite good at that kind of thing, has increasingly effectively referred developers to their WWDC videos. I'm also skeptical of the ability of the LLMs to ingest and properly synthesize that documentation - I'm willing to bet the answers from SO and Reddit are doing more heavy lifting on shaping the LLM's "answers" than you're hoping here.

There is nothing in my couple decades of programming or experience with LLMs that suggests to me that published documentation is going to be sufficient to let an LLM produce sufficient quality output without human synthesis somehwere in the loop.

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

#154

Post by Const-me »

stickfigure wrote: post #133
insane_dreamer wrote: post #120

The problem is eventually what are LLMs going’s to draw from? They’re not creating new information, just regurgitating and combining existing info. That’s why they perform so poorly on code for which there aren’t many many publicly available samples, SO/reddit answers etc.

> The problem is eventually what are LLMs going’s to draw from? Published documentation. I'm going to make up a number but I'll defend it: 90% of the information content of stackoverflow is regurgitated from some manual somewhere. The problem is that the specific information you're looking for in the relevant documentation is often hard to find, and even when found is often hard to read. LLMs are fantastic at reading…

That is only true for trivial questions.

I've answered dozens of questions on stackoverflow.com with tags like SIMD, SSE, AVX, NEON. Only a minority of these asked for a single SIMD instruction which does something specific. Usually people ask how to use the complete instruction set to accomplish something higher level.

Documentation alone doesn't answer questions like that, you need an expert who actually used that stuff.

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

#155

Post by elicksaur »

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…

On the other hand, I find it to be a fatal flaw that LLMs can’t say, “Hey you probably don’t actually want to do it that way.”

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

#156

Post by betaby »

tinco wrote: post #140
haolez wrote: post #72

Earlier quoted context omitted.

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…

I learned a bunch of programming languages on IRC, and the C and C++ communities on freenode were by far the most toxic I've encountered. Now that Rust is succesfully assimilating those communities, I have noticed the same toxicity on less well moderated forums, like the subreddit. The Discord luckily is still great. It's probably really important to separate the curmudgeons from the fresh initiates to provide an enj…

Not my experience. IRC was 'toxic' since forever, but that't not a toxicity, that's inability to read emotion through transactional plan text. Once one account that in the mental model IRC is just fine.

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

#157

Post by empath75 »

insane_dreamer wrote: post #120

The problem is eventually what are LLMs going’s to draw from? They’re not creating new information, just regurgitating and combining existing info. That’s why they perform so poorly on code for which there aren’t many many publicly available samples, SO/reddit answers etc.

AI companies are already paying humans to produce new data to train on and will continue to do that. There's also additional modalities -- they've already added text, video, and audio, and there's probably more possible. Right now almost all the content being fed into these AIs is stuff that humans can sense and understand, but why does it have to limit itself to that? There's probably all kinds of data types it could train on that could give it more knowledge about the world.

Even limiting yourself to code generation, there are going to be a lot of software developers employed to write or generate code examples and documentation just for AIs to ingest.

I think eventually AIs will begin coding in programming languages that are designed for AI to understand and work with and not for people to understand.

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

#158

Post by SunlitCat »

n_ary wrote: post #149
insane_dreamer wrote: post #120

The problem is eventually what are LLMs going’s to draw from? They’re not creating new information, just regurgitating and combining existing info. That’s why they perform so poorly on code for which there aren’t many many publicly available samples, SO/reddit answers etc.

LLMs show their limits as you try to ask something new(introduced in last 6-12 months) being not used. I was asking Claude and GPT4o about a new feature of go, it just gave me some old stuff from go docs. Then I went to go docs(official) and found what I was looking for anyways, the feature was released 2 major versions back, but somehow neither GPT4o nor claude know about this.

With GPT 4o I had some success pointing it to the current documentation of projects I needed and had it giving me current and actual answers.

Like "Help me to do this and that and use this list of internet resources to answer my questions"

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

#159

Post by epgui »

elicksaur wrote: post #151
epgui wrote: post #137

Earlier quoted context omitted.

In a very real sense, that’s also how human brains work.

This argument always conflates simple processes with complex ones. Humans can work with abstract concepts at a level LLMs currently can’t and don’t seem likely capable of. “True” and “False” are the best examples.

It doesn’t conflate anything though. It points to exactly that as a main difference (along with comparative functional neuroanatomy).

It’s helpful to realize the ways in which we do work the same way as AI, because it gives us perspective unto ourselves.

(I don’t follow regarding your true and false statement, and I don’t share your apparent pessimism about the fundamental limits of AI.)

Post reply on HN