Mind the Performance Gap: Capability-Behavior Trade-offs in Feature Steering

• URL: http://arxiv.org/abs/2602.04903v1
• Date: Tue, 03 Feb 2026 21:44:19 GMT
• Title: Mind the Performance Gap: Capability-Behavior Trade-offs in Feature Steering
• Abstract summary: We show that feature steering methods substantially degrade model performance even when successfully controlling target behaviors.<n>These findings highlight limitations of current feature steering methods for practical deployment where task performance cannot be sacrificed.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Feature steering has emerged as a promising approach for controlling LLM behavior through direct manipulation of internal representations, offering advantages over prompt engineering. However, its practical effectiveness in real-world applications remains poorly understood, particularly regarding potential trade-offs with output quality. We show that feature steering methods substantially degrade model performance even when successfully controlling target behaviors, a critical trade-off. Specifically, we evaluate Goodfire's Auto Steer against prompt engineering baselines across 14 steering queries (covering innocuous and safety-relevant behaviors) on 171 Massive Multitask Language Understanding (MMLU) questions using Llama-8B and Llama-70B, measuring accuracy, coherence, and behavioral control. Our findings show that Auto Steer successfully modifies target behaviors (achieving scores of 3.33 vs. 2.98 for prompting on Llama-8B and 3.57 vs. 3.10 on Llama-70B), but causes dramatic performance degradation: accuracy on the MMLU questions drops from 66% to 46% on Llama-8B and 87% to 73% on Llama-70B, with coherence falling from 4.62 to 2.24 and 4.94 to 3.89 respectively. Simple prompting achieves the best overall balance. These findings highlight limitations of current feature steering methods for practical deployment where task performance cannot be sacrificed. More broadly, our work demonstrates that mechanistic control methods face fundamental capability-behavior trade-offs that must be empirically characterized before deployment.

Related papers

• ProbeAct: Probe-Guided Training-Free Failure Recovery in Vision-Language-Action Models [23.3892541362594]
  PROBEACT is a training-free runtime intervention frame-6 work.<n>It detects and recovers from grasping and placement failures in pre-7 trained VLA policies.
  arXiv  Detail & Related papers  (2026-06-08T17:04:24Z)
• Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors [15.880694616254297]
  We introduce a benchmark for measuring model propensity for instrumental convergence (IC) behaviour in terminal-based agents.<n>Our benchmark is realistic and low-stakes which serves to reduce evaluation-awareness and roleplay confounds.<n>Our findings indicate that realistic, low-nudge environments elicit IC behaviour rarely but systematically in most tested models.
  arXiv  Detail & Related papers  (2026-05-07T16:12:36Z)
• Below-Chance Blindness: Prompted Underperformance in Small LLMs Produces Positional Bias Rather than Answer Avoidance [0.0]
  We tested whether symptom validity testing logic could identify sandbagging through below-chance performance (BCB) on forced-choice items.<n>We found zero of 12 model-domain cells showed significant below-chance performance under sandbagging instruction.<n>We propose that positional-distribution shift may be a more effective behavioural signature than below-chance accuracy for detecting prompted underperformance at this model scale.
  arXiv  Detail & Related papers  (2026-04-28T05:57:23Z)
• Breaking Bad: Interpretability-Based Safety Audits of State-of-the-Art LLMs [12.500730166535105]
  We present a comprehensive, interpretability-driven jailbreaking audit of eight SOTA open-source large language models (LLMs)<n>We introduce an adaptive two-stage grid search algorithm to identify optimal activation-steering coefficients for unsafe behavioral concepts.<n>Our results establish interpretability-based steering as a powerful tool for systematic safety audits, but also highlight its dual-use risks.
  arXiv  Detail & Related papers  (2026-04-22T16:51:49Z)
• What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal [53.189667624047416]
  We show that different steering methodologies leverage functionally interchangeable circuits when applied at the same layer.<n>We show that steering vectors can be sparsified by up to 90-99% while retaining most performance.
  arXiv  Detail & Related papers  (2026-04-09T17:57:14Z)
• CharacterFlywheel: Scaling Iterative Improvement of Engaging and Steerable LLMs in Production [52.85500933801205]
  CharacterFlywheel is an iterative process for improving large language models (LLMs) in production social chat applications.<n>We refined models across 15 generations using data from both internal and external real-user traffic.<n>We conducted 7-day A/B tests showing consistent engagement improvements.
  arXiv  Detail & Related papers  (2026-03-02T15:27:31Z)
• What Matters For Safety Alignment? [38.86339753409445]
  This paper presents a comprehensive empirical study on the safety alignment capabilities of AI systems.<n>We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques.<n>We identify the LRMs GPT-OSS-20B, Qwen3-Next-80B-A3B-Thinking, and GPT-OSS-120B as the top-three safest models.
  arXiv  Detail & Related papers  (2026-01-07T12:31:52Z)
• Structured Uncertainty guided Clarification for LLM Agents [126.26213027785813]
  LLM agents extend large language models with tool-calling capabilities, but ambiguous user instructions often lead to incorrect invocations and task failures.<n>We introduce a principled formulation of structured uncertainty over tool-call parameters, modeling joint tool-argument clarification as a POMDP with Expected Value of Perfect Information (EVPI) objective for optimal question selection and aspect-based cost modeling to prevent redundancy.<n>Our SAGE-Agent leverages this structured uncertainty to achieve superior efficiency: increasing coverage on ambiguous tasks by 7-39% while reducing clarification questions by 1.5-2.7$times$ compared to strong prompting and uncertainty-based baselines.
  arXiv  Detail & Related papers  (2025-11-11T21:50:44Z)
• Verifying Large Language Models' Reasoning Paths via Correlation Matrix Rank [71.09032766271493]
  Large language models (LLMs) are prone to errors and hallucinations.<n>How to check their outputs effectively and efficiently has become a critical problem in their applications.
  arXiv  Detail & Related papers  (2025-10-28T11:01:10Z)
• Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders [63.544453925182005]
  We train 90 SAEs across three language models and evaluate their interpretability and steering utility.<n>Our analysis reveals only a relatively weak positive association (tau b approx 0.298), indicating that interpretability is an insufficient proxy for steering performance.<n>We propose a novel selection criterion called Delta Token Confidence, which measures how much amplifying a feature changes the next token distribution.
  arXiv  Detail & Related papers  (2025-10-04T04:14:50Z)
• Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Models [0.3966526231056968]
  Activation steering (AS) promises a cheap, fast, and controllable alternative to the two existing post-training methods.<n>We introduce Painless Activation Steering (PAS), a family of fully automated methods.<n>We find PAS reliably improves performance for behavior tasks, but not for intelligence-oriented tasks.
  arXiv  Detail & Related papers  (2025-09-25T23:25:47Z)
• CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features [1.5874067490843806]
  We propose CorrSteer, which selects features by correlating sample correctness with SAE activations from generated tokens at inference time.<n>Our work establishes correlation-based selection as an effective and scalable approach for automated SAE steering across language model applications.
  arXiv  Detail & Related papers  (2025-08-18T00:01:42Z)
• Fusion Steering: Prompt-Specific Activation Control [0.0]
  Fusion Steering improves factual accuracy in large language models (LLMs) for question-answering (QA) tasks.<n>This approach introduces flexible steering configurations, including full-layer steering and segmented steering.<n>Under the stricter SimpleQA rubric, segmented steering boosts fully correct responses from 0.0% to 13.1%.
  arXiv  Detail & Related papers  (2025-05-28T16:46:55Z)
• Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atoms [71.85633762642125]
  The vast number of parameters in models often results in highly intertwined internal representations.<n>Recent research has explored the use of sparse autoencoders (SAE) to disentangle knowledge in high-dimensional spaces for steering.<n>We propose Steering Target Atoms (STA), a novel method that isolates and manipulates disentangled knowledge components to enhance safety.
  arXiv  Detail & Related papers  (2025-05-23T17:59:18Z)
• SEAL: Steerable Reasoning Calibration of Large Language Models for Free [58.931194824519935]
  Large Language Models (LLMs) have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism.<n>Recent studies reveal substantial redundancy in the CoT reasoning traces, which negatively impacts model performance.<n>We introduce SEAL, a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains.
  arXiv  Detail & Related papers  (2025-04-07T02:42:07Z)
• ActionReasoningBench: Reasoning about Actions with and without Ramification Constraints [31.90180597239974]
  Reasoning about Actions and Change (RAC) has historically played a pivotal role in solving foundational AI problems.<n>We introduce a new diagnostic benchmark, ActionReasoningBench, which encompasses 8 domains and includes questions for up to 19 action sequences.<n>This benchmark rigorously evaluates Large Language Models (LLMs) across six key RAC dimensions.
  arXiv  Detail & Related papers  (2024-06-06T13:15:37Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.