# Reasoning-Finetuning Repurposes Latent Representations in Base Models

arXiv: https://arxiv.org/abs/2507.12638 · HTML: https://mlsanity.com/paper/2507.12638/

## Key facts

- Cited by 26 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 12 references that are in the map.
- First posted on arXiv 16 Jul 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)

## Builds on (most cited first, 8 of 8)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Demystifying Long Chain-of-Thought Reasoning in LLMs](https://mlsanity.com/paper/2502.03373/) (2025)
- [Understanding Reasoning in Thinking Language Models via Steering Vectors](https://mlsanity.com/paper/2506.18167/) (2025)
- [I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2503.18878/) (2025)
- [Step Back to Leap Forward: Self-Backtracking for Boosting Reasoning of Language Models](https://mlsanity.com/paper/2502.04404/) (2025)
- [On the Emergence of Thinking in LLMs I: Searching for the Right Intuition](https://mlsanity.com/paper/2502.06773/) (2025)

## Built on by (most cited first, 26 of 26)

- [Intern-S1: A Scientific Multimodal Foundation Model](https://mlsanity.com/paper/2508.15763/) (2025)
- [Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought](https://mlsanity.com/paper/2603.05488/) (2026)
- [Base Models Know How to Reason, Thinking Models Learn When](https://mlsanity.com/paper/2510.07364/) (2025)
- [Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models](https://mlsanity.com/paper/2601.14004/) (2026)
- [Verifying Chain-of-Thought Reasoning via Its Computational Graph](https://mlsanity.com/paper/2510.09312/) (2025)
- [Reasoning Models Generate Societies of Thought](https://mlsanity.com/paper/2601.10825/) (2026)
- [Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectors](https://mlsanity.com/paper/2509.06608/) (2025)
- [The Value Axis: Language Models Encode Whether They're on the Right Track](https://mlsanity.com/paper/2606.17056/) (2026)
- [Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failures](https://mlsanity.com/paper/2601.19928/) (2026)
- [Steerable but Not Decodable: Function Vectors Operate Beyond the Logit Lens](https://mlsanity.com/paper/2604.02608/) (2026)
- [How's it going? Reinforcement learning in language models recruits a functional welfare axis](https://mlsanity.com/paper/2605.30232/) (2026)
- [Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Process](https://mlsanity.com/paper/2512.23988/) (2025)
- [Rank-1 LoRAs Encode Interpretable Reasoning Signals](https://mlsanity.com/paper/2511.06739/) (2025)
- [How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signals](https://mlsanity.com/paper/2604.22271/) (2026)
- [Consolidating Rewarded Perturbations for LLM Post-Training](https://mlsanity.com/paper/2605.31494/) (2026)
- [Convergent World Representations and Divergent Tasks](https://mlsanity.com/paper/2602.00533/) (2026)
- [Reasoning Can Be Restored by Correcting a Few Decision Tokens](https://mlsanity.com/paper/2605.16874/) (2026)
- [Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal](https://mlsanity.com/paper/2606.12360/) (2026)
- [Transcoder Adapters for Reasoning-Model Diffing](https://mlsanity.com/paper/2602.20904/) (2026)
- [Fluid Reasoning Representations](https://mlsanity.com/paper/2602.04843/) (2026)
- [MoRFI: Monotonic Sparse Autoencoder Feature Identification](https://mlsanity.com/paper/2604.26866/) (2026)
- [PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning](https://mlsanity.com/paper/2605.23074/) (2026)
- [Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs](https://mlsanity.com/paper/2606.00726/) (2026)
- [Reasoning Fine-Tuning Induces Persistent Latent Policy States](https://mlsanity.com/paper/2607.18532/) (2026)
- [Context-Grounding Gains Are Mediated by Pre-existing Machinery: Auditing GRPO, SFT, and DPO](https://mlsanity.com/paper/2609.00925/) (2026)
- [The Thinking Leakage in Reinforcement Learning for LLMs](https://mlsanity.com/paper/2610.14270/) (2026)

## Abstract (arXiv)

Backtracking, an emergent behavior elicited by reasoning fine-tuning, has been shown to be a key mechanism in reasoning models' enhanced capabilities. Prior work has succeeded in manipulating this behavior via steering vectors, but the underlying mechanism remains poorly understood. In this work, we show that the emergence of backtracking in DeepSeek-R1-Distill-Llama-8B is in part driven by a repurposed direction already present in base model activations. Specifically, we identify a direction in base Llama-3.1-8B's residual stream which systematically induces backtracking when used to steer the distilled reasoning model, and find that the effects of steering with this direction cannot be trivially explained by token-level attributes. We further find that this direction does not induce backtracking in the base model, suggesting that the reasoning finetuning process repurposes pre-existing representations to form new behavioral circuits. Additionally, we hypothesize that this direction is one of several which may work together to mediate backtracking. Our findings offer a compelling picture that reasoning-finetuned models repurpose pre-existing base model representations, rather than learn new capabilities from scratch.