[ExI] 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*
<https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/>

* John K Clark    See what's on my list at  Extropolis
<https://groups.google.com/g/extropolis>*

4y9
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20260909/42184462/attachment-0001.htm>