Abstract:Large language models can generate long chain-of-thought (CoT) reasoning, yet prior work suggests that CoT can be post-hoc rationalization rather than a faithful reflection of the computation through explicitly designed settings. In this work, we go further and propose a True Thinking Score (TTS) to quantify the causal contribution of each step in CoT to the model's final prediction in realistic reasoning problems. Across eleven models ranging from 1.5B to 1.1T parameters on common reasoning benchmarks, we find that CoTs often interleave true-thinking steps, which causally affect the final answer, with decorative-thinking steps, which appear useful but have little causal influence; Such decorative steps remain prevalent even for frontier models: Over 30% of steps in Kimi-K2.6 are decorative on MATH with TTS <= 0.005. Furthermore, TTS enables effective CoT pruning: removing 50% of CoT steps with the lowest TTS can largely maintain the performance. Self-training on these pruned CoTs reduces reasoning length by 66% while preserving performance on Nemotron3-Nano-30B. Finally, we provide a mechanistic analysis showing that LLMs can be steered in the latent space to engage or disengage with reasoning steps. Overall, our results reveal that frontier LLMs often verbalize reasoning steps that are not causally used, challenging both the efficiency and the trustworthiness of CoT.
From: Jiachen Zhao [view email]
[v1]
Tue, 28 Oct 2025 20:14:02 UTC (1,522 KB)
[v2]
Fri, 30 Jan 2026 07:56:32 UTC (1,619 KB)
[v3]
Mon, 27 Apr 2026 00:14:34 UTC (1,618 KB)
[v4]
Tue, 26 May 2026 22:58:09 UTC (2,176 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。