Beyond a Global Norm: Personalizing Toxicity Sensitivity in Language Models Without Retraining

A study from arXiv (2607.23175v1) presents the first comparative evaluation of training-free methods for aligning language models to user-specific toxicity sensitivities, finding that all methods reduce alignment error by 28-47% but reveal a fundamental trade-off between alignment effectiveness, personalization, and language quality.

arXiv:2607.23175v1 Announce Type: new Abstract: Reducing toxicity is often framed as a global alignment problem, yet perceptions of harmful language are subjective and context-dependent. We present the first comparative evaluation of training-free methods for aligning language generation to user-specific toxicity sensitivities across three inference-time intervention stages: pre-decoding prompt conditioning and rewriting , in-decoding token, logit, and representation steering , and post-decoding candidate re-ranking . Evaluated against toxicity sensitivity targets derived from the PRISM dataset, all methods reduce alignment error by 28-47%. However, the results reveal a fundamental trade-off between alignment effectiveness, personalization, and general language quality, showing how toxicity sensitivity alignment is an inherently multi-objective problem.