Temporal Predictors of Outcome in Reasoning Language Models
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
The chain-of-thought (CoT) paradigm uses the elicitation of step-by-step rationales as a proxy for reasoning, gradually refining the model's latent representation of a solution. However, it remains unclear just how early a Large Language Model (LLM) internally commits to an eventual outcome. We probe this by training linear classifiers on hidden states after the first t reasoning tokens, showing that eventual correctness is highly predictable after only a few tokens, even when longer outputs are needed to reach a definite answer. We show that, for harder questions, a drop in predictive accuracy highlights a selection artifact: hard items are disproportionately represented in long CoTs. Overall, our results imply that for reasoning models, internal self-assessment of success tends to emerge after only a few tokens, with implications for interpretability and for inference-time control.
Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 8 of its 10 references that are in the map. First posted on arXiv 3 Nov 2025.
Builds on
8 of 10 references.
References · 8
- 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
- 2022Large Language Models are Zero-Shot Reasonerscited by 4,374
- 2022Least-to-Most Prompting Enables Complex Reasoning in Large Language Modelscited by 1,407
- 2024OpenAI o1 System Cardcited by 1,244
- 2025No Answer Needed: Predicting LLM Answer Accuracy from Question-Only Linear Probescited by 38
- 2025Early Stopping Chain-of-thoughts in Large Language Modelscited by 20
- 2025Inside-Out: Hidden Factual Knowledge in LLMscited by 7
Built on by
Citing papers per year: 2026: 6.
6 papers.
- 2026Triggering Chain-of-Thought via Latent Feature Interventions in Large Language Modelscited by 11
- 2026Breaking the Solver Bottleneck: Training Task Generators at the Learnable Frontiercited by 3
- 2026Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generationcited by 2
- 2026Latent Programming Horizons in Coding Agentscited by 1
- 2026It's the Problem, Not the Path: Budget and Difficulty Confounds in LLM Reasoning Trajectoriescited by 1
- 2026Temporal Preference Concepts and their Functions in a Large Language Modelcited by 0
- 2026It's the Problem, Not the Path: Budget and Difficulty Confounds in LLM Reasoning Trajectories3 Sep 2026
- 2026Latent Programming Horizons in Coding Agents6 Jul 2026
- 2026Breaking the Solver Bottleneck: Training Task Generators at the Learnable Frontier10 Jun 2026
- 2026Temporal Preference Concepts and their Functions in a Large Language Model11 May 2026
- 2026Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generation16 Feb 2026
- 2026Triggering Chain-of-Thought via Latent Feature Interventions in Large Language Models12 Jan 2026
Similar, not cited
Close in wording, no citation either way
- 2025Real-Time Progress Prediction in Reasoning Language Models
- 2025Knowing Before Saying: LLM Representations Encode Information About Chain-of-Thought Success Before Completion
- 2025Lost at the Beginning of Reasoning
- 2026Beyond the Commitment Boundary: Probing Epiphenomenal Chain-of-Thought in Large Reasoning Models
- 2025Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models
- 2026ReEfBench: Quantifying the Reasoning Efficiency of LLMs
- 2026Large Language Models Decide Early and Explain Later
- 2026State of Thought Enables Endogenous Reasoning