ARBOR: Detecting reasoning failures in long chains of thought
In brief
- ARBOR models reasoning trajectories as a state tree to track failures.
- It distinguishes three failure types with specific mechanisms and costs.
- Enables real-time corrections or integration into training data.
ARBOR is a framework for attributing reasoning failures in large reasoning models. It represents each reasoning trajectory as a tree of states, enabling the identification and classification of failures (overthinking, underthinking, unsupported deductions) by analyzing step-state relationships. The tool suggests ways to integrate these analyses into existing correction mechanisms.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error