Joined 14 years ago
Bringing up OpenAI's lawsuits is irrelevant to whether these proofs hold. And calling a release that includes Lean formalizations a "demonstration of power" gets it backwards. Machine checkable proofs are the least "trust me" form of mathematics there is.
There are fair criticisms here. Not every result is formalized, the model can't be reproduced by outsiders, and the massive dump strains review capacity. Those are reasons to demand full formalization, open access to the methods, and help funding human review. They aren't reasons to dismiss correct mathematics or to tell people to stop working on hard problems.
> https://docs.anthropic.com/en/docs/claude-code/sub-agents
I sort of feel if you're only figuring this out now you've been willfully/woefully ignorant.
Bringing up OpenAI's lawsuits is irrelevant to whether these proofs hold. And calling a release that includes Lean formalizations a "demonstration of power" gets it backwards. Machine checkable proofs are the least "trust me" form of mathematics there is.
There are fair criticisms here. Not every result is formalized, the model can't be reproduced by outsiders, and the massive dump strains review capacity. Those are reasons to demand full formalization, open access to the methods, and help funding human review. They aren't reasons to dismiss correct mathematics or to tell people to stop working on hard problems.