A Course Correction in Steerability Evaluation: Revealing Miscalibration and Side Effects in LLMs

• URL: http://arxiv.org/abs/2505.23816v1
• Date: Tue, 27 May 2025 21:29:52 GMT
• Title: A Course Correction in Steerability Evaluation: Revealing Miscalibration and Side Effects in LLMs
• Authors: Trenton Chang, Tobias Schnabel, Adith Swaminathan, Jenna Wiens,
• Abstract summary: It remains unclear whether large language models can produce outputs aligned with a broad variety of user goals.<n> Interventions to improve steerability, such as prompt engineering, have varying effectiveness.<n>Even strong LLMs struggle with steerability, and existing alignment strategies may be insufficient.
• Score: 14.334903198382287
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Despite advances in large language models (LLMs) on reasoning and instruction-following benchmarks, it remains unclear whether they can reliably produce outputs aligned with a broad variety of user goals, a concept we refer to as steerability. The abundance of methods proposed to modify LLM behavior makes it unclear whether current LLMs are already steerable, or require further intervention. In particular, LLMs may exhibit (i) poor coverage, where rare user goals are underrepresented; (ii) miscalibration, where models overshoot requests; and (iii) side effects, where changes to one dimension of text inadvertently affect others. To systematically evaluate these failures, we introduce a framework based on a multi-dimensional goal space that models user goals and LLM outputs as vectors with dimensions corresponding to text attributes (e.g., reading difficulty). Applied to a text-rewriting task, we find that current LLMs struggle with steerability, as side effects are persistent. Interventions to improve steerability, such as prompt engineering, best-of-$N$ sampling, and reinforcement learning fine-tuning, have varying effectiveness, yet side effects remain problematic. Our findings suggest that even strong LLMs struggle with steerability, and existing alignment strategies may be insufficient. We open-source our steerability evaluation framework at https://github.com/MLD3/steerability.

Related papers

• Enhancing LLM Robustness to Perturbed Instructions: An Empirical Study [8.827173113748701]
  We study character- and word-level edits of task-specific instructions, which substantially degrade downstream performance.<n>We find that, on average, self-denoising achieves substantially higher performance gains than alternative strategies.
  arXiv  Detail & Related papers  (2025-04-03T16:17:56Z)
• What You See Is Not Always What You Get: An Empirical Study of Code Comprehension by Large Language Models [0.5735035463793009]
  We investigate the vulnerability of large language models (LLMs) to imperceptible attacks, where hidden character manipulation in source code misleads LLMs' behaviour while remaining undetectable to human reviewers.<n>These attacks include coding reordering, invisible coding characters, code deletions, and code homoglyphs.<n>Our findings confirm the susceptibility of LLMs to imperceptible coding character attacks, while different LLMs present different negative correlations between perturbation magnitude and performance.
  arXiv  Detail & Related papers  (2024-12-11T04:52:41Z)
• CogSteer: Cognition-Inspired Selective Layer Intervention for Efficiently Steering Large Language Models [37.476241509187304]
  Large Language Models (LLMs) achieve remarkable performance through pretraining on extensive data.<n>The lack of interpretability in their underlying mechanisms limits the ability to effectively steer LLMs for specific applications.<n>We introduce an efficient selective layer intervention based on prominent parameter-efficient fine-tuning methods.
  arXiv  Detail & Related papers  (2024-10-23T09:40:15Z)
• Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making [85.24399869971236]
  We aim to evaluate Large Language Models (LLMs) for embodied decision making.<n>Existing evaluations tend to rely solely on a final success rate.<n>We propose a generalized interface (Embodied Agent Interface) that supports the formalization of various types of tasks.
  arXiv  Detail & Related papers  (2024-10-09T17:59:00Z)
• CLAMBER: A Benchmark of Identifying and Clarifying Ambiguous Information Needs in Large Language Models [60.59638232596912]
  We introduce CLAMBER, a benchmark for evaluating large language models (LLMs) Building upon the taxonomy, we construct 12K high-quality data to assess the strengths, weaknesses, and potential risks of various off-the-shelf LLMs. Our findings indicate the limited practical utility of current LLMs in identifying and clarifying ambiguous user queries.
  arXiv  Detail & Related papers  (2024-05-20T14:34:01Z)
• A Causal Explainable Guardrails for Large Language Models [29.441292837667415]
  Large Language Models (LLMs) have shown impressive performance in natural language tasks, but their outputs can exhibit undesirable attributes or biases. We propose LLMGuardrail, a novel framework that incorporates causal analysis and adversarial learning to obtain unbiased steering representations.
  arXiv  Detail & Related papers  (2024-05-07T09:55:05Z)
• Evaluation and Improvement of Fault Detection for Large Language Models [30.760472387136954]
  This paper investigates the effectiveness of existing fault detection methods for large language models (LLMs) We propose textbfMuCS, a prompt textbfMutation-based prediction textbfConfidence textbfSmoothing framework to boost the fault detection capability of existing methods.
  arXiv  Detail & Related papers  (2024-04-14T07:06:12Z)
• InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance [56.184255657175335]
  We develop textbfInferAligner, a novel inference-time alignment method that utilizes cross-model guidance for harmlessness alignment. Experimental results show that our method can be very effectively applied to domain-specific models in finance, medicine, and mathematics. It significantly diminishes the Attack Success Rate (ASR) of both harmful instructions and jailbreak attacks, while maintaining almost unchanged performance in downstream tasks.
  arXiv  Detail & Related papers  (2024-01-20T10:41:03Z)
• Machine Vision Therapy: Multimodal Large Language Models Can Enhance Visual Robustness via Denoising In-Context Learning [67.0609518552321]
  We propose to conduct Machine Vision Therapy which aims to rectify the noisy predictions from vision models. By fine-tuning with the denoised labels, the learning model performance can be boosted in an unsupervised manner.
  arXiv  Detail & Related papers  (2023-12-05T07:29:14Z)
• Are Large Language Models Really Robust to Word-Level Perturbations? [68.60618778027694]
  We propose a novel rational evaluation approach that leverages pre-trained reward models as diagnostic tools. Longer conversations manifest the comprehensive grasp of language models in terms of their proficiency in understanding questions. Our results demonstrate that LLMs frequently exhibit vulnerability to word-level perturbations that are commonplace in daily language usage.
  arXiv  Detail & Related papers  (2023-09-20T09:23:46Z)
• Prompting Large Language Models for Counterfactual Generation: An Empirical Study [13.506528217009507]
  Large language models (LLMs) have made remarkable progress in a wide range of natural language understanding and generation tasks. We present a comprehensive evaluation framework on various types of NLU tasks, which covers all key factors in determining LLMs' capability of generating counterfactuals.
  arXiv  Detail & Related papers  (2023-05-24T06:44:32Z)
• Editing Large Language Models: Problems, Methods, and Opportunities [51.903537096207]
  This paper embarks on a deep exploration of the problems, methods, and opportunities related to model editing for LLMs. We provide an exhaustive overview of the task definition and challenges associated with model editing, along with an in-depth empirical analysis of the most progressive methods currently at our disposal. Our objective is to provide valuable insights into the effectiveness and feasibility of each editing technique, thereby assisting the community in making informed decisions on the selection of the most appropriate method for a specific task or context.
  arXiv  Detail & Related papers  (2023-05-22T16:00:00Z)

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.