OpenAI Uses Simulated Users for Multi-Turn LLM Evaluation
OpenAI built an evaluation system with hundreds of detailed simulated user personas to directly compare assistant models in multi-turn conversations, using a pretrained base model for natural user messages filtered by a specially trained navigator model.
2026-09-01 ~ 2026-09-01 · 2 related posts
- LLM Eval Uses Simulated Users for Multi-turn Comparison — _ddjohnson · 2026-09-01
- Using Base Models for More Natural User Simulations — _ddjohnson · 2026-09-01