Beyond a Global Norm: Personalizing Toxicity Sensitivity in Language Models Without Retraining
Quick Answer
This study evaluates training-free methods for personalizing toxicity sensitivity in language models, achieving a 28-47% reduction in alignment error.
Quick Take
The methods are assessed across pre-decoding, in-decoding, and post-decoding stages, highlighting a trade-off between alignment effectiveness, personalization, and overall language quality.
Key Points
- Evaluated methods reduce alignment error by 28-47% based on PRISM dataset targets.
- Three intervention stages include pre-decoding, in-decoding, and post-decoding.
- Findings reveal a trade-off between personalization and language quality.
- Study emphasizes the subjective nature of toxicity perception.
- No retraining of models is required for these methods.
Paper Resources
Source Excerpt
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 tox
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.