AI Paper Atlas

Research map / Large language models / Alignment and preferences

HumanLM: Simulating Users with State Alignment Beats Response Imitation

Shirley Wu, Evelyn Choi, Arpandeep Khatua, Zhanghan Wang, Joy He-Yueya, Tharindu Cyril Weerasooriya, Wei Wei, Diyi Yang, Jure Leskovec, James Zou

ICML 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 0 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
replace SFT-think with HumanLM latent-state alignmentHumanLM · Humanual-Avg.response alignment score: 8.6 -> 13.2Improved
replace standard GRPO-think with HumanLMHumanLM · Humanual-Avg.response alignment score: 10.4 -> 13.2Improved
add reasoning (think) to SFTSFT · Humanual-Avg.response alignment score: 6.5 -> 8.6Improved
add reasoning (think) to Qwen3-8bQwen3-8b · Humanual-Avg.response alignment score: 9.5 -> 8.4Hurt
add reasoning (think) to GRPOGRPO · Humanual-Avg.response alignment score: 10.3 -> 10.4No clear effect
use GRPO on alignment score instead of SFTQwen3-8b · Humanual-Emailresponse alignment score: 3.94 -> 4.78Improved
use SFT imitation instead of base modelQwen3-8b · Humanual-Avg.response alignment score: 9.5 -> 6.5Hurt
train HumanLM vs GRPO-think on state alignment explorationHumanLM · Humanual-Opinion trainingspan of state alignment score: 23% higherImproved
train HumanLM vs GRPO-think on response alignment explorationHumanLM · Humanual-Opinion trainingspan of response alignment score: 104% higherImproved
HumanLM vs Qwen3-8b-think embedding similarityHumanLM · Humanual-Avg.embedding similarity: 42.5 -> 45.7Improved
HumanLM vs GRPO-think in human study win rateHumanLM · Real-time user studywin rate on overall similarity (%): 27.9% -> 41.4%Improved
HumanLM vs Qwen3-8b-think in human humanlikenessHumanLM · Real-time user studyshare rated quite natural or above (%): 72.1% -> 76.6%Improved

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