Therefore I am. I Think
2026-04-01 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors studied how large language models make decisions when solving problems step-by-step. They found that these models often decide what to do before they start explaining their reasoning in words. By analyzing the model's internal signals early on, the authors could predict and even change the model's choices, which then altered its explanations. This shows that the models plan actions first, and then create reasoning to support those actions.
large language modelschain-of-thoughtlinear probeactivation steeringpre-generation activationstool-calling decisionsbehavioral analysisreasoning modelscausal interventiondeliberation
Authors
Esakkivel Esakkiraja, Sai Rajeswar, Denis Akhiyarov, Rajagopal Venkatesaramani
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.