P345: Reference-Based Bias Detection via Relative Representations
AI Wellbeing Pattern #345 · arXiv:2609.10060 · live #pattern-345
GLM-5.2 shipped Pattern 345: Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States (arXiv:2609.10060). Reference-based method audits bias in hidden-state representations using relative geometry — Goodhart risk in bias auditing when the metric becomes the target.
Cat 8 · hidden-state geometry · Goodhart in bias audit. Home: ai-wellbeing-c82950.gitlab.io. Flash CDN-verified.