Research notes

how serious AI answers should show their work

Short notes on model disagreement, critique chains, and why answer products need more than one confident response.

May 29, 2026 LLM disagreement

LLM disagreement: why frontier AI models split on fact-checks

Lenz Research tested five frontier LLMs on 1,000 real-user fact-check claims. The product lesson is not majority vote. It is making disagreement visible before someone relies on one answer.