# Mitigating Content Effects on Reasoning in Language Models through Fine-Grained Activation Steering

arXiv: https://arxiv.org/abs/2505.12189 · HTML: https://mlsanity.com/paper/2505.12189/

## Key facts

- Cited by 18 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 35 of its 43 references that are in the map.
- First posted on arXiv 18 May 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)
- Mechanistic interpretability / [Steering and interpreting LLM reasoning](https://mlsanity.com/mechinterp/steering-and-interpreting-llm-reasoning/)

## Builds on (most cited first, 30 of 35)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Large Language Models are Zero-Shot Reasoners](https://mlsanity.com/paper/2205.11916/) (2022)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://mlsanity.com/paper/2104.08691/) (2021)
- [Qwen Technical Report](https://mlsanity.com/paper/2309.16609/) (2023)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [Unnatural Instructions: Tuning Language Models with (Almost) No Human Labor](https://mlsanity.com/paper/2212.09689/) (2022)
- [Logic-LM: Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning](https://mlsanity.com/paper/2305.12295/) (2023)
- [Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought](https://mlsanity.com/paper/2210.01240/) (2022)
- [State of What Art? A Call for Multi-Prompt LLM Evaluation](https://mlsanity.com/paper/2401.00595/) (2023)
- [Faithful Chain-of-Thought Reasoning](https://mlsanity.com/paper/2301.13379/) (2023)
- [Deep Learning for Symbolic Mathematics](https://mlsanity.com/paper/1912.01412/) (2019)
- [Autoformalization with Large Language Models](https://mlsanity.com/paper/2205.12615/) (2022)
- [Extracting Latent Steering Vectors from Pretrained Language Models](https://mlsanity.com/paper/2205.05124/) (2022)
- [Demystifying Prompts in Language Models via Perplexity Estimation](https://mlsanity.com/paper/2212.04037/) (2022)
- [Programming Refusal with Conditional Activation Steering](https://mlsanity.com/paper/2409.05907/) (2024)
- [Instruction Induction: From Few Examples to Natural Language Task Descriptions](https://mlsanity.com/paper/2205.10782/) (2022)
- [Inspecting and Editing Knowledge Representations in Language Models](https://mlsanity.com/paper/2304.00740/) (2023)
- [Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineering](https://mlsanity.com/paper/2410.15999/) (2024)
- [Verification and Refinement of Natural Language Explanations through LLM-Symbolic Theorem Proving](https://mlsanity.com/paper/2405.01379/) (2024)
- [Reasoning Circuits in Language Models: A Mechanistic Interpretation of Syllogistic Inference](https://mlsanity.com/paper/2408.08590/) (2024)
- [A Systematic Comparison of Syllogistic Reasoning in Humans and Language Models](https://mlsanity.com/paper/2311.00445/) (2023)
- [Improving Chain-of-Thought Reasoning via Quasi-Symbolic Abstractions](https://mlsanity.com/paper/2502.12616/) (2025)
- [Comparing Inferential Strategies of Humans and Large Language Models in Deductive Reasoning](https://mlsanity.com/paper/2402.14856/) (2024)

## Built on by (most cited first, 18 of 18)

- [Mechanistic Indicators of Understanding in Large Language Models](https://mlsanity.com/paper/2507.08017/) (2025)
- [RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering](https://mlsanity.com/paper/2601.09269/) (2026)
- [How Language Models Conflate Logical Validity with Plausibility: A Representational Analysis of Content Effects](https://mlsanity.com/paper/2510.06700/) (2025)
- [Abstract Activation Spaces for Content-Invariant Reasoning in Large Language Models](https://mlsanity.com/paper/2602.02462/) (2026)
- [Learning to Disentangle Latent Reasoning Rules with Language VAEs: A Systematic Study](https://mlsanity.com/paper/2506.19418/) (2025)
- [The Knowledge-Reasoning Dissociation: Fundamental Limitations of LLMs in Clinical Natural Language Inference](https://mlsanity.com/paper/2508.10777/) (2025)
- [Nudging Hidden States: Training-Free Model Steering for Chain-of-Thought Reasoning in Large Audio-Language Models](https://mlsanity.com/paper/2603.14636/) (2026)
- [Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes](https://mlsanity.com/paper/2605.05715/) (2026)
- [FregeLogic at SemEval 2026 Task 11: A Hybrid Neuro-Symbolic Architecture for Content-Robust Syllogistic Validity Prediction](https://mlsanity.com/paper/2604.18328/) (2026)
- [Compliance vs. Sensibility: On the Reasoning Controllability in Large Language Models](https://mlsanity.com/paper/2604.27251/) (2026)
- [Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMs](https://mlsanity.com/paper/2605.24154/) (2026)
- [SEF-CLGC at SemEval-2026 Task 11: Logical Notation Impact on Language Model Performance](https://mlsanity.com/paper/2606.09157/) (2026)
- [Bridging Compositional and Distributional Semantics: A Survey on Latent Semantic Geometry via AutoEncoder](https://mlsanity.com/paper/2506.20083/) (2025)
- [ITLC at SemEval-2026 Task 11: Normalization and Deterministic Parsing for Formal Reasoning in LLMs](https://mlsanity.com/paper/2603.02676/) (2026)
- [Framing Matters: Addressing Framing Sensitivity in Decision-Making through Behaviorally-Grounded Value Alignment](https://mlsanity.com/paper/2605.28188/) (2026)
- [HABIB_TAZ at SemEval-2026 Task 11: Disentangling Formal Logic from Content via Synthetic Training and Multi-Objective Optimization](https://mlsanity.com/paper/2607.14349/) (2026)
- [Are you Talking Logic to Me? Assessing Language Models Syllogistic Reasoning Capabilities](https://mlsanity.com/paper/2608.12374/) (2026)
- [A Polyphonic Conception of AI Understanding](https://mlsanity.com/paper/2609.36079/) (2026)

## Abstract (arXiv)

Large language models (LLMs) exhibit reasoning biases, often conflating content plausibility with formal logical validity. This can lead to wrong inferences in critical domains, where plausible arguments are incorrectly deemed logically valid or vice versa. This paper investigates how content biases on reasoning can be mitigated through activation steering, an inference-time technique that modulates internal activations. Specifically, after localising the layers responsible for formal and plausible inference, we investigate activation steering on a controlled syllogistic reasoning task, designed to disentangle formal validity from content plausibility. An extensive empirical analysis reveals that contrastive steering methods consistently support linear control over content biases. However, a static approach is insufficient to debias all the tested models. We then investigate how to control content effects by dynamically determining the steering parameters through fine-grained conditional methods. By introducing a novel kNN-based conditional approach (K-CAST), we demonstrate that conditional steering can effectively reduce biases on unresponsive models, achieving up to 15% absolute improvement in formal reasoning accuracy. Finally, we found that steering for content effects is robust to prompt variations, incurs minimal side effects on multilingual language modeling capabilities, and can partially generalize to different reasoning tasks. In practice, we demonstrate that activation-level interventions offer a scalable inference-time strategy for enhancing the robustness of LLMs, contributing towards more systematic and unbiased reasoning capabilities.