Refining user representations improves the fidelity of AI user simulation

The authors train a representation generator with hindsight distillation to outperform GPT-5.5 on conversation reconstruction across seven datasets.

Big Tech
Shuhaib Mehri · Philippe Laban · Michel Galley · Dilek Tür · Jennifer Neville · Jianfeng Gao

University of Illinois Urbana-Champaign · Microsoft Research

Research Digest··3 min read
The authors identify natural-language user representations as an underexplored bottleneck in user simulation and formulate conversation reconstruction to evaluate them.

In user simulation, a simulator is conditioned on a user representation, a natural-language description meant to reproduce a particular user's behavior.

§