TheoremSmith: Building Long-Horizon Proof Tasks to Train LLMs
·8 min read
Author: @charles1huang Table of Contents Why Improve Formal Verification? Verifiable Domains Are Low-Hanging Sowing Seeds Verifying the Agents Synthesis The Factory Results Limitations Future Work
Author: @charles1huang Table of Contents Why Improve Formal Verification? Verifiable Domains Are Low-Hanging Sowing Seeds Verifying the Agents Synthesis The Factory Results Limitations Future Work TL;DR: Making models better at formal verification will help push the boundaries of math research and significantly reduce the number of bugs in our code. Recently, training models on RL environments (tasks paired with verifiers that reward correct solutions) has driven big gains in model performance,
52
0
12
0