CRoS: Rewarding CoT causality boosts LLM faithfulness +13.8% on Gemma-3-4B
mohitban47 · x · 2026-11-03
A UNC PhD team proposes CRoS (Causality Reward on Scrambled CoTs) to improve LLM reasoning accuracy and CoT faithfulness together. Outcome rewards raise accuracy but can push models toward shortcuts and post-hoc rationalization; CRoS instead rewards causal contribution of the CoT — if reasoning causally drives the answer, scrambling its steps should change the answer probability. On Gemma-3-4B, CRoS lifts CoT faithfulness by 13.8% while also improving accuracy and reasoning quality, mitigating the accuracy–faithfulness trade-off.
More from Research
- Claude writes neural net weights from scratch: 959 programs, 100% accuracy on 10 algorithm tasks — ChengleiSi · 2026-11-03
- Critic says viral AI interp demo was 'a bit of a bait': the prompt hinted the answer — mayfer · 2026-11-03
- Qwen got stuck in a loop: Astra fixed it by ablating a 'looping direction' from weights — burny_tech · 2026-11-03
- Recursive GPT-5-Mini calls often beat a single GPT-5 call for less money — lateinteraction · 2026-11-03
- Same Agent Reimplements Liquid in Rust vs C: Rust Uses 42% Fewer Tokens, C Runs 1.7x Faster — antirez · 2026-11-03
- After viral 'fly plays Beat Saber', researchers call for shared whole-brain emulation benchmark — rishabh16_ · 2026-11-03