New method trains generative agents with step-level human preference data

Researchers have developed a new method for training generative agents in social simulations by collecting step-level human preference data. This approach involves an interactive simulation interface to gather over 57,000 fine-grained annotations on agent decision trajectories. By applying supervised fine-tuning and direct preference optimization to open-weight language models using this data, the study demonstrates consistent improvements in simulation fidelity, coordination, and the overall quality of agent behavior. AI

IMPACT This research could lead to more sophisticated and human-aligned AI agents in simulations, improving their ability to plan and act over long horizons.

RANK_REASON The cluster contains a research paper detailing a novel method for training AI agents. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Direct Preference Optimization: Your Language Model is Secretly a Reward Model
• Generative Agents: Interactive Simulacra of Human Behavior
• Gotit.pub
• Hugging Face
• ScienceCast
• Social Simulations for Border Security

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →