These would seem contradictory. If you really think that both are true and Altman knows it, then you're saying he's a hype man lying for regulatory capture. And to some extent he definitely is overblowing the danger for his own gain.
I really doubt they are a dead end though, we've barely started to explore what they can do. There's a lot more that can be extracted from existing datasets, multimodality, gains in GPU power to wait for, fine tunes for use cases that don't even have datasets yet, etc. Just the absolute mountain of things we've learned since LLama came out are enough to warrant base model retrains.
Only if you believe that LLM is a synonym for AI, which OpenAI doesn't.
The things Altman have said seem entirely compatible with "the danger to humanity is ahead of us, not here and now", although in part that's because of the effort put into making GPT-4 refuse to write propaganda for Al Quaida, as per the red team safety report they published at the same time as releasing the model.
Other people are very concerned with here-and-now harms from AI, but that's stuff like "AI perpetuates existing stereotypes" and "when the AI reaches a bad decision, who do you turn to to get it overturned?" and "can we, like, not put autonomous tasers onto the Boston Dynamics Spot dogs we're using as cheap police substitutes?"
> LLMs seem like a dead end
These would seem contradictory. If you really think that both are true and Altman knows it, then you're saying he's a hype man lying for regulatory capture. And to some extent he definitely is overblowing the danger for his own gain.
I really doubt they are a dead end though, we've barely started to explore what they can do. There's a lot more that can be extracted from existing datasets, multimodality, gains in GPU power to wait for, fine tunes for use cases that don't even have datasets yet, etc. Just the absolute mountain of things we've learned since LLama came out are enough to warrant base model retrains.