Classic and High-Value AI Papers

Goal: master durable papers first, then follow new frontier work.

Core fundamentals

• Attention Is All You Need (2017)
• BERT (2018)
• GPT-3 (2020)
• InstructGPT (2022)
• Chain-of-Thought Prompting (2022)

Alignment and safety

• Constitutional AI (2022)
• Direct Preference Optimization (2023)
• RLAIF related work collection

Agents and tool use

• ReAct (2022)
• Toolformer (2023)
• Self-Refine (2023)

Multimodal and generation

• CLIP (2021)
• DALL·E 2 (2022)
• Segment Anything (2023)

Reading order

1. Start with abstract, figures, and key results.
2. Read methods next, then training details and appendix.
3. Write one-page notes per paper: problem, method, evidence, limitations, transferability.