# Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task

Authors: Alicia Curth, Rachel Lawrence, Sushrut Karmalkar, Niranjani Prasad.

Studies when transformer layers combine family relationships to answer multi-hop questions. Uses logit-lens probes and causal activation patching to compare depth use before and after task fine-tuning.

First public: 2026-04-14.
ArXiv initial submission: 2026-04-14.

[Original paper](https://arxiv.org/abs/2604.12426) · [PDF](https://arxiv.org/pdf/2604.12426) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2604.12426/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2604.12426/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2604.12426/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2604.12426/record.json)

## 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 0.15.2; literature checked through 2026-11-03.