Hacker News Reader

Comment by rockemsockem | Hacker News Reader

I really feel like you can't have used any advanced LLMs if you legitimately think the out put "perfect white noise". The results that you can get from an LLM like GPT-4 are incredibly useful and are providing an enormous amount of value to lots of people. It isn't just for generating phony information to spread or having it do your work for you.

I get the most value out of asking for examples of things or asking for basic explanations or intuitions about things. And I get so much value from this that I really think the printing press is the most apt comparison.

View on HN

The problem is advanced LLMs are controlled by large corporations. Powerful local models exist (in part thanks to Meta's generosity oddly enough) and they're close to GPT-3.5, but GPT-4 is far ahead of them and by the time other models reach to that point whatever OpenAI or Antropic, Meta etc. have developed behind closed doors could be significantly better. In that case open models will be restricted to niche uses and most people will use the latest model from a giant corp.

So it is possible that LLMs will centralize the production and dissemination of knowledge, which is the opposite of what people think the printing press did. I hope I'm wrong and open models can challenge/overtake state of the art models developed by tech giants, that would be amazing.

Precisely. I spent weeks learning about cybersecurity when GPT-4 first came out, as I could finally ask as many stupid questions as I liked, get detailed examples and use-cases for different attacks and defenses, and generally actually learn how the internet around me works.

Now it refuses, because OpenAI's morals apparently don't include spreading openly available knowledge about how to defend yourself.

Scary. I have also been using it to generate useful political critiques (given a particular theoretical tradition, some style notes, and specific articles to critique, it's actually excitingly good). What if OpenAI decides that's a threat? What reason do we have to think that a powerful institution would not take this course of action, in the cold light of history?

how do you know what you learnt wasn't completely made up gibberish?

The same way you know that the things you learn from a person isn't made up gibberish: You see how well it explains a scenario, how well it lines up with your knowledge and experience, and you sample parts to verify.

how do you know they didn't learn from the garbage generator too?

You are literally describing the fundamental problem of truth in philosophy and acting as if it's different because a computer is involved at one step in the chain.

Because it works.

What you say is not in conflict with AI-generated content being white noise. Even if you find some piece of AI-generated content useful, it is still white noise if it is merely combining pieces of information found in its dataset and the result is posted online or published elsewhere. There is no signal being added in that process, and it pollutes the space of content. Humans are also prone to doing this, but with the help of AI, it becomes a much larger issue.

"Signal" would mean new data, which is by definition not possible via LLMs trained on publicly available content, since that means the data is already out there, or new and meaningful ideas or innovations beyond just combining existing material. I have not seen LLMs accomplish the latter. I consider it at least possible that they are capable of such a feat, but even then the relevant question would be how often they produce such things compared to just rearranging existing content. Is the proportion high enough that unleashing floods of AI-generated content everywhere would not lower the signal-to-noise ratio from the pre-AI situation?

Reply on news.ycombinator.com