# Present but Hard to Read: Probing Latent Reasoning with Minimal Pairs

Authors: Aur Shalev Merin.

Uses controlled prompt changes and interventions to separate information present in hidden states from what a probe can read and what a model actually uses. Tests copied digits and unwritten arithmetic intermediates in a looped model and a standard transformer.

First public: 2026-10-20.
ArXiv initial submission: 2026-10-20.

[Original paper](https://arxiv.org/abs/2610.28498) · [PDF](https://arxiv.org/pdf/2610.28498) · [HTML](https://arxiv.org/html/2610.28498v2) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2610.28498/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2610.28498/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2610.28498/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2610.28498/record.json)
[Code](https://github.com/Aur-Shalev-Merin/present-but-hard-to-read)

## Approaches

- [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.