TensorX
返回文献探索

Paper · arXiv 2604.01202

Therefore I am. I Think

Esakkivel Esakkiraja, Sai Rajeswar, Denis Akhiyarov, Rajagopal Venkatesaramani

34 upvotesApril 2, 2026arXiv 预印本
AI 摘要

Reasoning models appear to encode action choices before beginning textual deliberation, as evidenced by early decision detection and activation steering effects.

chain-of-thoughtlinear probeactivation steeringtool-calling decisionspre-generation activationsdeliberationbehavioral analysis

Abstract

We consider the question: when a large language reasoning model makes a choice, did it think first and then decide to, or decide first and then think? In this paper, we present evidence that detectable, early-encoded decisions shape chain-of-thought in reasoning models. Specifically, we show that a simple linear probe successfully decodes tool-calling decisions from pre-generation activations with very high confidence, and in some cases, even before a single reasoning token is produced. Activation steering supports this causally: perturbing the decision direction leads to inflated deliberation, and flips behavior in many examples (between 7 - 79% depending on model and benchmark). We also show through behavioral analysis that, when steering changes the decision, the chain-of-thought process often rationalizes the flip rather than resisting it. Together, these results suggest that reasoning models can encode action choices before they begin to deliberate in text.

北京市昌平区探索星信息技术及软件开发工作室

京ICP备2026059466号
Therefore I am. I Think | TensorX