MAS.S62 AI4AI | Schedule

** Exact topics and schedule subject to change, based on student interests and course discussions. **

Date	Topics	Readings
9/9	Week 1 Course introduction [slides] Basics of LLMs and foundation models AI agents and self-evolution
9/16	Week 2 Context evolution 1 [slides] [colab] Prompt optimization Memory, tools, knowledge	Large Language Models Are Human-Level Prompt Engineers TextGrad: Automatic "Differentiation" via Text DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines Reflexion: Language Agents with Verbal Reinforcement Learning Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory
9/23	Week 3 Context evolution 2 [slides] LLMs as optimizers LLMs for discovery	FunSearch: Mathematical discoveries from program search with large language models Large Language Models as Optimizers AlphaEvolve: A coding agent for scientific and algorithmic discovery
9/30	Week 4 Project proposal presentations
10/7	Week 5 Harness evolution 1 [slides] Harness components Agent memory Tool and skill evolution	Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents Recursive Language Models Context Language Models MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-horizon Agents Reflexion: Language Agents with Verbal Reinforcement Learning Voyager: An Open-Ended Embodied Agent with Large Language Models Meta-Harness: End-to-End Optimization of Model Harnesses Continual Harness: Online Adaptation for Self-Improving Foundation Agents HarnessTax: How Much Does the Harness Matter for Coding Agents? Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents
10/14	Week 6 Harness evolution 2 [slides] [harness tutorial] Multi-agent systemss Multi-agent evolution Harness engineering tutorial	CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discovery Towards a Science of Scaling Agent Systems Why Do Multi-Agent LLM Systems Fail?
10/21	Week 7 Model evolution 1 [slides] Continual learning Distillation Test-time training
10/28	Week 8 Project midterm presentations
11/4	Week 9 Model evolution 2 [slides] Self-play approaches
11/11	Week 10 Environment evolution 1 [slides] Synthetic data and tasks Environment generation
11/18	Week 11 Environment evolution 2 [slides] World models Evolving evaluation and rewards
11/25	Week 12 Future topics [slides] AI for scientific discovery Multimodal agents
12/2	Week 13 Project final presentations
12/9	Week 14 Future topics [slides] Embodied harnesses Safe evolution