Paper proves transformers are inherently succinct and that basic verification problems like emptiness/equivalence are EXPSPACE-complete, making formal LLM verification provably intractable.

Research on regret minimization algorithms that adapt to opponent behavior in repeated games.