CRoS Rewards Causal Chain-of-Thought, Boosting Faithfulness by 13.8%

UNC researchers propose CRoS, which rewards causal reliance on chain-of-thought by using scrambled CoTs, improving both accuracy and faithfulness (up 13.8%) in LLM reasoning.

2026-11-03 ~ 2026-11-03 · 2 related posts

• CRoS: rewarding causality on scrambled CoTs boosts faithfulness 13.8% on Gemma-3-4B — mohitban47 · 2026-11-03

1 near-duplicate retellings: mohitban47