Recursive self improvement and GPT-7

[John Clark]

Anthropic released Claude Fable 5 to the general public about two weeks ago and everybody was amazed, it was by far the most advanced AI that anybody had ever seen. Then about one week ago OpenAI released GPT-6 to the general public and it blew Claude Fable 5 out of the water, it was by far the most advanced AI that anybody had ever seen. Then just yesterday OpenAI announced that one of their AIs solved the million dollar millennium mathematical prize involving the Navier-Stokes Equation which describes fluid flow; the AI proved that even though the equation is based entirely on Newtonian physics it nevertheless produces mathematical singularities when things get small. Human mathematicians have been trying to prove that, or trying to prove that doing that was impossible, since 1822 when the Navier-Stokes Equation was discovered, but they had been unsuccessful in doing either.  

The interesting thing was the AI that OpenAI used to create the biggest mathematical breakthrough in 25 years was NOT GPT-6, it was another unreleased AI (internal code name "Bell") that the company said had "significantly" greater abilities then GPT-6. Bell generated 10,000 agents, each investigating a different way of attacking the problem, and after thinking about it for 88 hours it solved a problem that humans had been trying to solve for over 200 years. 

It took 33 months to go from GPT-3 to GPT-4, but now with recursive self improvement .... well... we live in interesting times.  

AI Has Solved One of Math’s $1 Million Millennium Prize Problems

 John K Clark    See what's on my list at  Extropolis

4y9

[John Clark]

This is the sort of thing I was referring to. And this is why I can't get all hot and bothered over deficit spending, illegal immigration, climate change, or men in women's sports:

We need to talk about this...

John K Clark