arXiv AI By Seunghee Koh, Sungjae Choi, Minchan Kwon, Sunghyun Baek, Junmo Kim

</think> Doesn't Stop Reasoning: Analysis of Spurious CoT Termination

Read the original on arXiv AI →

The paper investigates a training‑free early‑exit technique that inserts an end‑of‑think (EoT) token to terminate chain‑of‑thought (CoT) reasoning in large reasoning models. It finds that the injected EoT often fails to cleanly switch the model from reasoning to answering, leading to continued reasoning‑like generation—termed spurious CoT termination—whose length scales with the amount of reasoning saved. By increasing attention to the EoT token through Exit‑token Attention Biasing (EAB), the authors reduce spurious termination and shorten the answering phase across multiple models and benchmarks.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

arXiv AI
2d ago

When Can Large Reasoning Models Save Thinking? Mechanistic Analysis of Behavioral Divergence in Reasoning

The paper investigates why large reasoning models (LRMs) often continue to think even when prompted to stop, a phenomenon called "Still-thinking". By examining confidence at the thinking-termination boundary, internal attention divergences, and attention allocation across prompt segments, the authors find that high perplexity and greater attention to the original question correlate with continued thinking. They propose an attention‑intervention method that suppresses explicit reasoning, which reduces inefficiency but also lowers accuracy, underscoring a trade‑off between instruction compliance, inference speed, and correctness.

By Rongzhi Zhu, Yi Liu, Jiancheng Wang, Xiangyu Liu, Zequn Sun, Yiwei Wang, Yu Deng, Zijian Zhou, Wei Hu
arXiv Computation and Language
Aug 28

TRACES: Tagging Reasoning Steps for Adaptive Cost-Efficient Early-Stopping

TRACES (Tagging Reasoning Steps for Adaptive Cost‑Efficient Early‑Stopping) is a lightweight framework that tags reasoning steps of large‑language models in real time, enabling adaptive, cost‑efficient early stopping during inference. By monitoring the types of steps generated, the method identifies when models shift their reasoning after arriving at a correct answer, allowing for interpretable stopping criteria. Experiments on mathematical reasoning benchmarks (MATH500, GSM8K, AIME) and knowledge benchmarks (MMLU, GPQA) show token reductions of 20–50% while preserving accuracy, with more conservative thresholds needed for harder tasks such as BeyondAIME and IMO AnswerBench.

By Yannis Belkhiter, Seshu Tirupathi, Giulio Zizzo, John D. Kelleher