CASE

Researchers propose CASE, a framework combining training-time causal alignment and inference-time structural enforcement to improve chain-of-thought faithfulness in large language models. Experiments …