Tim Pearce

Papers

Total Citations

H-Index

About

Tim Pearce is a researcher whose work sits at the intersection of machine learning, reinforcement learning, and robotics, with a particular focus on understanding the scaling laws that govern intelligent agent behavior. His most notable contribution, the 2024 paper "Scaling Laws for Pre-training Agents and World Models," provides a foundational framework showing that the performance of embodied agents—from robots to video game characters—improves predictably with increases in model parameters, dataset size, and compute. This work, which has already garnered early citations, mirrors the scaling insights that have revolutionized large language models, but applies them to the more complex domain of agentic, interactive systems. By demonstrating that generative pre-training on offline behavioral data yields consistent performance gains, Pearce has helped chart a path toward more capable and generalist AI agents. His research is particularly impactful for students and researchers working at the frontier of foundation models for robotics and game AI, offering both a theoretical grounding and a practical roadmap for building increasingly powerful embodied systems.