ArgGYM: A Procedural, Engine-Verified Benchmark for Structured Defeasible Reasoning
Recent progress in large language model reasoning has been driven by benchmarks and reinforcement learning environments with automatically verifiable rewards, particularly in mathematics, code, and formal logic.
arXiv - cs.AI2 min read