Think-to-Talk or Talk-to-Think? When LLMs Come Up with an Answer in Multi-Step Reasoning
Kudo, Keito, Aoki, Yoichi, Kuribayashi, Tatsuki, Sone, Shusaku, Taniguchi, Masaya, Brassard, Ana, Sakaguchi, Keisuke, Inui, Kentaro
–arXiv.org Artificial Intelligence
This study investigates the internal reasoning mechanism of language models during symbolic multi-step reasoning, motivated by the question of whether chain-of-thought (CoT) outputs are faithful to the model's internals. Specifically, we inspect when they internally determine their answers, particularly before or after CoT begins, to determine whether models follow a post-hoc "think-to-talk" mode or a step-by-step "talk-to-think" mode of explanation. Through causal probing experiments in controlled arithmetic reasoning tasks, we found systematic internal reasoning patterns across models; for example, simple subproblems are solved before CoT begins, and more complicated multi-hop calculations are performed during CoT.
arXiv.org Artificial Intelligence
Dec-1-2024
- Country:
- Asia (0.47)
- North America > United States
- Minnesota (0.14)
- Genre:
- Research Report > New Finding (0.93)
- Technology: