Egregoros

Signal feed

Timeline

Post

Remote status

Context

1
The problem is the "big AI labs" (OpenAI, Anthropic) each thought AI was gonna become a monopoly and they would be on top of it - but it didn't, it became a commodity and no one is on top.

So the only way they can even HOPE to recover their investor money is to pivot into applied AI, which is where little random startups are making bank. The problem is that's a hard problem, lots of startups are trying to do it, only a few are gonna be successful - and it's most likely that internal teams inside of OpenAI or Anthropic are NOT gonna be amongst the few...
@cjd @caekislove It's a hard problem too, because it looks a lot like a management problem (Just make the AI guy do the thing!) but without domain expertise, it's quite challenging to scope the model to do the right thing; and synthetic domain expertise (skills, memories) is really, really hard! It's like training- super easy to brick your model with incorrect inputs, or sway the solution into something you REALLY DO NOT want.

Replies

0

Fetching replies…