# Not All LLM Reasoning is Visible in the Chain-of-Thought

Authors: Vatsal Baherwani, Tom Goldstein, Ashwinee Panda.

Tests whether fixed filler tokens improve reasoning without exposing intermediate steps in text. Finds task- and model-dependent effects, studies their internal mechanisms, and reports limits of training models to exploit them.

First public: 2026-07-24.
ArXiv initial submission: 2026-07-24.

[Original paper](https://arxiv.org/abs/2607.22925) · [PDF](https://arxiv.org/pdf/2607.22925) · [HTML](https://arxiv.org/html/2607.22925v2) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2607.22925/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2607.22925/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2607.22925/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2607.22925/record.json)

## Approaches

- [Thinking tokens](https://latent-implicit-thinking.github.io/topics/thinking-tokens/): Give a model extra tokens for computation, such as pauses, filler symbols, or learned planning tokens. These tokens need not spell out the solution.
- [Analysis and evaluation](https://latent-implicit-thinking.github.io/topics/implicit-reasoning-evaluation/): Measure how well these methods work, investigate what their internal computation does, and identify where they fail. This includes experiments, studies of model internals, and mathematical results.

Reading-list version 0.11.0; literature checked through 2026-10-26.