Artificial Intelligence (AI) has made remarkable progress through increasingly capable foundation models trained on large-scale data. However, today's systems remain largely static after training, requiring substantial human effort to improve their capabilities, acquire new knowledge, and adapt to changing environments. This course introduces the emerging field of self-evolving AI: AI systems that can improve themselves through experience, feedback, memory, interaction, and exploration to continually adapt and optimize their behavior over long time horizons. We will cover methods for (1) context evolution through memory, retrieval, tool use, and long-context reasoning, (2) harness evolution through feedback, search, workflow optimization, and multi-agent systems, (3) model evolution through self-play, distillation, test-time training, and continual learning, (4) task evolution through generating increasingly complex synthetic tasks, simulations, evaluations, and environments.

Through lectures, readings, discussions, homeworks, and a significant research component, this course will develop knowledge of recent technical achievements in self-evolving AI, critical thinking skills and intuitions in designing self-evolving AI, and a deeper understanding of the AI research process. Students will complete hands-on assignments on self-evolving AI, culminating in a novel research project.

• Time: Wednesday 2:00pm-4:00pm
• Location: MIT Media Lab E15-359