Counting on Thinking: Tracing Evidence Integration in Language Models
cs.LG
Submitted: 2026-09-26
Updated: 2026-09-26
Terminology
Sources
- Why do LLMs attend to the first token?
- Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
- From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step
- Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
- Contextual Position Encoding: Learning to Count What's Important
- Chain of Thought Empowers Transformers to Solve Inherently Serial Problems
- The Expressive Power of Transformers with Chain of Thought
- LLMs Are In-Context Bandit Reinforcement Learners
- NVIDIA Nemotron 3: Efficient and Open Intelligence
- Show Your Work: Scratchpads for Intermediate Computation with Language Models
- In-context Learning and Induction Heads
- OpenAI o1 System Card
- ICLR: In-Context Learning of Representations
- Larger language models do in-context learning differently
- Large language models reorganize representational geometry during in-context learning
- When Can Transformers Count to n?
- Distilling System 2 into System 1
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks