AI Paper Atlas

Research map / Large language models / Alignment and preferences

Flipping the Dialogue: Training and Evaluating User Language Models

Tarek Naous, Philippe Laban, Wei Xu, Jennifer Neville

ICLR 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 47 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use trained UserLM-8b instead of GPT-4o as user simulatorGPT-4o assistant · GSM8K and HumanEval simulated conversationsassistant score (%): 74.6 -> 57.4Hurt
use UserLM-8b instead of GPT-4o as simulatoruser simulation · PRISM dialogue terminationF1: 1.38 -> 63.54Improved
train a user LM (UserLM-8b) instead of prompting GPT-4ouser simulation · NaturalnessPangram human-likelihood (%): 3.31 -> 80.21Improved
train UserLM-8b instead of prompting GPT-4ouser simulation · First-turn diversity1-gram Jaccard (%): 74.42 -> 94.55Improved
train UserLM-8b instead of prompting GPT-4ouser simulation · Intent decomposition1-gram overlap (%): 7.68 -> 2.69Improved
train UserLM-8b instead of prompting GPT-4ouser simulation · User role adherenceaccuracy (%): 38.85 -> 93.95Improved
train UserLM-8b instead of prompting GPT-4ouser simulation · Intent adherenceaccuracy (%): 70.95 -> 94.65Improved
scale UserLM from 1b to 8bUserLM · PRISM perplexity with intentperplexity: 18.58 -> 14.92Improved
scale the prompted assistant from GPT-4o-mini to GPT-4oprompted user simulator · Six intrinsic metricsmetric winsNo clear effect
train UserLM from instruction-tuned checkpoint instead of base (ablation)UserLM · PRISM perplexityperplexityHurt
start training from 1b instead of 8b instruction-tuned assistant (ablation)UserLM from instruct checkpoint · PRISM perplexityperplexity: 27.25 -> 18.45Improved
train with intent conditioning (ablation)UserLM · PRISM perplexityperplexityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy