I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.
ATLAS: Autoformalized Textbook Library At ScaleBoth tackle scaling formal mathematical proof at massive scale, one through generative-verifier RL test-time scaling and the other through a growing library of autoformalized textbooks.
Supported by
Material Discovery Bench: LLM Research BenchmarkMaxProof's approach to scaling generative-verifier RL for mathematical proof is analogous to verifiable scientific discovery benchmarks — both represent LLMs being evaluated on ground-truth-constrained expert tasks
What Sort of Maths Are LLMs Good At?MaxProof's work on scaling mathematical proof generation with RL is a concrete example of the kind of mathematical capability the new item is analyzing and categorizing
Leiden Declaration on Artificial Intelligence and MathematicsMaxProof's scaling of mathematical proof via generative-verifier RL is precisely the kind of AI-mathematics integration the Leiden Declaration would address as a research priority and governance concern.
Generative Recursive Reasoning (GRAM)GRAM's recursive refinement of a persistent latent state offers a reasoning mechanism that could underlie the population-level test-time scaling MaxProof uses for generating and verifying proofs.
Related
Kimi Vendor VerifierKimi Vendor Verifier and MaxProof both center on verifier mechanisms that validate correctness of model-generated outputs before acceptance.
Improving Gpt 5 6 Sol In ChatgptBoth concern scaling mathematical and problem-solving capabilities in frontier models — MaxProof's approach to mathematical proof via RL and OpenAI's GPT improvements represent parallel efforts toward stronger AI reasoning
mattwood.fyi