Sam Devlin

Papers

Total Citations

H-Index

About

Sam Devlin is a leading researcher at the intersection of artificial intelligence, generative modeling, and sequential decision-making. His primary research areas include imitation learning, reinforcement learning, and the scaling of embodied agents. Devlin’s most notable contribution is pioneering the use of diffusion models—typically associated with text-to-image generation—as observation-to-action models for imitating complex, stochastic human behavior in sequential environments. His 2023 paper on this topic, “Imitating Human Behaviour with Diffusion Models,” has already garnered 23 citations, reflecting its immediate impact on the field. More recently, Devlin has explored the scaling laws governing pre-trained agents and world models, demonstrating in his 2024 work that increasing model parameters, dataset size, and compute can significantly enhance performance across domains like robotics and video games. This research provides critical insights into how generative learning objectives on offline datasets can be used to model agent behavior effectively. Devlin’s work is shaping the future of AI agents capable of nuanced, human-like interaction.