Back to feed
arXiv cs.CL
arXiv cs.CL
8/3/2026
Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges

Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges

Short summary

Chain-of-Models (CoM) is an automated audit pipeline where a second LLM inspects the first model's reasoning trace before producing a final judgment. Across 9 models, 4 biases, and 4 datasets, the best auditor is bias-specific: GPT-4o handles bandwagon/authority/distraction while GLM-5 handles sycophancy. A per-bias auditor selection rule achieves 0.884 accuracy versus 0.805 for no-audit baseline.

  • •CoM uses a second LLM to audit the first model's reasoning trace for cognitive biases
  • •Best auditor is bias-specific: GPT-4o for bandwagon/authority/distraction, GLM-5 for sycophancy
  • •Per-bias auditor selection reaches 0.884 accuracy vs 0.805 no-audit baseline

Generated with AI, which can make mistakes.

#research-breakthrough#ai-agents#ai-tools
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

Interventional Grounding Audits: Black-Box Premise-Dependency Tests for LLM Chain-of-Thought via Predicate Substitution
arXiv CS.AI
Beyond Shapley: An Influence-Based Data Auditing Pipeline for LLM Alignment and Evaluation
arXiv cs.LG
The Formalism Trap: LLM-as-a-Judge Evaluators Confuse Procedural Structure with Semantic Truth
arXiv cs.CL
AI Evals, Part 4: LLM-as-Judge, Done Right
Dev.to