HumanLM

HumanLM: Simulating Users with State Alignment Beats Response Imitation

We build user simulators that accurately reflect real users by generating natural-language latent states aligned with ground-truth responses. Fig 1. HumanLM generates responses by first producing latent states (stance, emotion, communication style) that align with ground-truth user behavior, then synthesizing responses from these aligned states. Abstract Large Language Models are increasingly used to simulate how specific users respond to any context, enabling more user-centric applications. However, existing user simulators mostly imitate surface-level patterns and language styles, which…

saved by

• Laerdon Kim

related reading

• Quantifying the Utility of User Simulators for Building Collaborative LLM Assistantsarxiv.org
• user_interactions.pdfself-distillation.github.io
• Lost in Simulation: LLM-Simulated Users are Unreliable Proxies for Human Users in Agentic Evaluationsarxiv.org
• [2508.19227] Generative Interfaces for Language Modelsarxiv.org
• HumanLM: Simulating Users with State Alignment Beats Response Imitationarxiv.org
• Behavior Latticing: Inferring User Motivations from Unstructured Interactionsarxiv.org
• [2304.03442] Generative Agents: Interactive Simulacra of Human Behaviorarxiv.org
• [2304.03442] Generative Agents: Interactive Simulacra of Human Behaviorarxiv.org
• Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activationstransformer-circuits.pub
• Training and Evaluating User Language Modelsarxiv.org
• [2606.06614] Re-Centering Humans in LLM Personalizationarxiv.org
• Mechanistically Eliciting Latent Behaviors in Language Models — AI Alignment Forumalignmentforum.org