SwtCyber

And what's interesting is that formalization doesn't just catch missing cases, it also forces you to examine all the assumptions that everyone has been taking for granted

I think you're basically right about the end goal yet the gap between a textbook proof and a Lean proof is much bigger than it looks

> Verifying that actions adhere to instruction correctly and will continue to do so in the future, even when improvising autonomously. This may be an open problem that has no solution, and is maybe the missing piece that would allow people to entrust AI to act on their behalf

Instead of trying to achieve perfect instruction adherence, it’s much safer to sandbox the execution environment at the infrastructure layer. If the container only has write access to a single scratch bucket, whatever weird improvisation the model comes up with is at least safely contained

I never really thought about how unreliable early mechanical clocks must have been....

The smaller providers on OpenRouter are probably popping champagne rn. With a framework like this in hand, the barrier to entry for LLM hosting drops even lower, so expect another wave of api price cuts for all no-frontier models

I love that we've reached a point where you can put an entire N64 level on your dining table and physically walk around it.

Exactly. I suspect the problem isn't just the amount of training data, but how much of the relevant information is missing from it.

I think the cross-language comparison would be interesting, although the bouba/kiki effect seems remarkably consistent across human languages already