There’s no reason to assume it’s the best solution. It might be the case that a better tokenization scheme is needed for math, reasoning, video, etc models.
Yeah great way to become dependent on tech stacks backdoored by different big tech companies from another country, or the same ones who move to different countries.
This is a European initiative, we are already backdoored by other companies from other countries. I'd rather work with a company I can take to court/that is governed by policies I can vote for or against.
Yes, the European Union, who so earnestly considered writing a law that forces tech companies to build in backdoors to E2EE because they think only they could get in.
You want the European Parliament and every European country to have direct access to everything you type and say? Because that’s where this ends.
A SAT solver without any preprocessing won't be competitive with SoTA SAT solver.
CDCL is core to the problem, but it is not sufficient. You even have SAT solvers like CryptoMiniSAT that try to detect clauses that encode xor gates so they can use Gaussian Elimination.
This is also true of ILP solvers. Simplex + Branch & Cut is elegant. But that's not how you get to the top.
I don’t see the advantage of this approach. You end up with a worse and less reliable circuit just to avoid the single chemical step.
reply