Really important admission from Altman here: > I think the way that most of us define AGI, we’re still missing something quite important,... one big one is … this is not a model that continuously learns as it’s deployed from things it finds www.theguardian.com/technology/2...
That’s vital, because that gap is a fundamental part of how LLMs work. They’re trained, but they can’t learn. It’s a one-time process. You can add context, you can have them talk to themselves, but that’s not learning. It’s just applying the same knowledge to a more complex (and expensive) problem.
OpenAI’s valuation (and ability to raise more cash to cover their losses) depends on their potential to build a human-replacing AGI. What we’ve got so far is useful in a lot of cases, but nowhere near good enough to justify the trillions being poured into it (1% of US GDP!)
Altman is admitting that he’s reached the end of the LLM road. If LLMs aren’t going to be enough then either: 1. They’ve got the next thing under the table, a genuine scientific breakthrough. Or, 2. The industry needs to rescale their ambition to “making some knowledge workers 10% more efficient”
If it’s option 1 then they need to get on with it. OpenAI is loosing $40bn a year. If they’re sitting on the most valuable technology in the world then they need to put up or shut up.
If it’s option 2 then that’s still amazing. The man who made every software dev in the world a little bit more efficient would still be a very successful man. But he wouldn’t be culturally significant in the way AI is at the moment, and it wouldn’t justify the the current crazy valuations.