@adrusi@shlevy Same could be said about Julia/Mandlebrot set. Took half a century for them to develop computers powerful enough to realize how interesting they are.
Mathematical formulation is even simpler for them as well.
@vamsiprithamp In time, everything is on the chopping block. Robotics will be solved sooner than later, converting everything into a cognitive task.
There will be adjustments, but much more good will come from this than bad.
@alz_zyd_ The relevant comparison is not their salaries but the opportunity cost of incentivizing extremely talented individuals away from working on problems to benefit humanity.
@teortaxesTex@growing_daniel The reference class includes Taiwan as one obvious example.
If you include Singapore and HK, you get a swath of sino nations all of which are govern better than the CCP.
@vamsiprithamp like I would think these would be an order of magnitude higher. Maybe that's my own biases, or maybe LLM capabilities are just very spiky.
> The total number of tokens needed to find solutions to these problems would cost roughly $2,000 at Sol API rates.
Surprising to me that these problems are so "cheap" to solve relative to how much we spend on software engineering projects.
yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model.
We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more.
More thoughts here: https://t.co/8SjXONeh38
@vamsiprithamp Well "proofs" seems a bit overpowered for a collision list. And I wonder if even then it is so precise based on the patient and other context.
Definitely a problem that needs better fixing though.
@vamsiprithamp I think RL is definitely interesting. Although there is also the argument that so much of the published literature is wrong that it may hurt more than it helps π
@vamsiprithamp I'd be curious to know if there is indeed anything that is amenable to a proof. All I've experienced is dissapointment π
Perhaps there is room to formally verify software for things like pace makers?