| Week | Topic | Readings / Materials | Discussion Goal |
|---|---|---|---|
| 1 | LLM foundations and scaling | Stanford CS336; Stanford CS229 | Build common vocabulary for LLM systems, training, and evaluation. |
| 2 | Machine learning theory for LLMs | Columbia COMS 6998; intro slides | Connect theory tools to modern model behavior. |
| 3 | Post-training and preference optimization | DPO, RLHF, GRPO, reward models, verifiers, post-training harnesses | Clarify what has been validated at meaningful scale. |
| 4 | OPD and industry-verified practice | External notes and technical reports to be added | Understand what is correct, reliable, and scalable in practice. |
| 5 | Reasoning, verification, and process supervision | Verifier-guided learning, reasoning diagnostics, process supervision papers | Study how to evaluate and improve reasoning beyond benchmark chasing. |
| 6 | Kimi, Hunyuan, Qwen, and frontier model systems | Moonshot AI; Tencent Hunyuan; Qwen Blog | Read industrial technical reports as research objects. |
| 7 | Featured reading: Ruiqi Zhang et al. | arXiv:2306.09927 | Discuss representation, reasoning, and why this work matters. |
| 8 | Project formation | Course notes and shared reading summaries | Form project directions from accumulated discussions. |
This schedule is under development. Members are encouraged to suggest readings, seminar pages, replication targets, and technical notes.