[Hot Take] LLMs are getting boring/overhyped

[OmegAI]

Hey folks!

I just want to have a friendly discussion here not trying to be negative nancy or anything like that.

I’ve been using chatgpt for a while now and it is very obviously gaining a huge amount of popularity in such a small amount of time. I am a student studying ML and I find the evolvement of field to be deeply fascinating both to work on and observe. I think that much of the work that OpenAI has done with DALL-E and their hide and seek were super fascinating. Self play is truly incredible and I remember when the agents (seekers) were able to find and exploit a bug in the physics engine to surf on top of boxes. Generally speaking, I think that most of the work in RL (Hide&Seek/AlphaGo/etc) is greatly under appreciated.

Nowadays, especially after AI has become mainstream, the only thing people seem to care about are LLMs. Of course GPTs are SOTA, and while I don’t think the field of NLP is entirely boring and while I don’t doubt that later versions are going to become exponentially better, I feel like they have been too overhyped and that a lot of other fantastic work in AI has been overshadowed.

Just a thought, I hope that the future is more exciting than just trying to scale LLMs. I think it will be. I hope it will be. And I think we are definitely missing things. RL could play a huge part in the future, maybe other ideas we haven’t discovered yet as well. I think the best future is that which is the most exciting. Curious about what you all think about this.

[curt.kennedy]

@OmegAI

Nowadays, especially after AI has become mainstream, the only thing people seem to care about are LLMs. Of course GPTs are SOTA, and while I don’t think the field of NLP is entirely boring and while I don’t doubt that later versions are going to become exponentially better, I feel like they have been too overhyped and that a lot of other fantastic work in AI has been overshadowed.

I wouldn’t say LLM’s are the only models people care about. What about self driving cars? Or self navigating drones?

While LLM’s have the spotlight, what work do you think is overshadowed? I agree a lot is overshadowed, especially with all the research in AI created daily.

[elmstedt]

@OmegAI

I feel like they have been too overhyped and that a lot of other fantastic work in AI has been overshadowed.

Overshadowed or not, that other work continues apace and continues to get funding.

I don’t think there’s a real problem here, in fact one could argue that the halo-effect of having any AI product storm the cultural zeitgeist the way ChatGPT has is a big tide lifting a lot of other AI boats.

[_j]

It seems OpenAI is now the ChatGPT company. Forgotten ML:

gym.openai.com has a security policy called HTTP Strict Transport Security (HSTS), which means that Firefox can only connect to it securely. You can’t add an exception to visit this site.

(getting served by wrong Azure cert now. Archive.org mirror - one of the few archived OpenAI sites not blocked by robots.txt, or useless because of exclusively javascript page rendering))