r2m-ai/load-bearing-cot

Perturb-and-continue causal tests of chain-of-thought: does the written reasoning actually constrain the answer? Code, data, and probes for "From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought" (TMLR 2026).

语言：Python · GitHub 星标：189 · Fork：25

在 GitHub 上查看原仓库