Reference reading

Source register

The papers, documentation and courses our lessons cite, by reference number.

1. S01 NIST AI Risk Management Framework

   Organizational risk-management reference for AIE/AIF

2. S02 Python CSV documentation

   Standard-library CSV reader and text-conversion behavior

3. S03 scikit-learn: common pitfalls

   Training/test isolation and preprocessing discipline

4. S04 Google Machine Learning Crash Course

   Optional foundational ML reading

5. S05 Hugging Face LLM course

   Further study in LLMs and the Transformers ecosystem

6. S06 Anthropic: building effective agents

   Distinguishing workflows and model-directed agents

7. S07 Ollama chat API

   Optional adapter's /api/chat, messages, JSON format, non-streamed response

8. S08 MCP: build a server

   Official route for a future protocol-compliant implementation

9. S09 Wei et al., chain-of-thought prompting

   RSN-02 few-shot reasoning prompts and scale effects

10. S10 Kojima et al., zero-shot reasoners

    RSN-02 "Let's think step by step"

11. S11 Turpin et al., unfaithful chain-of-thought explanations

    RSN-03 faithfulness

12. S12 Snell et al., scaling test-time compute

    RSN-04 serial and parallel inference compute

13. S13 Wang et al., self-consistency

    RSN-05 sampling and majority voting

14. S14 Cobbe et al., training verifiers

    RSN-06 verifiers for sampled solutions

15. S15 Lightman et al., let's verify step by step

    RSN-06 outcome and process supervision

16. S16 DeepSeek-AI, DeepSeek-R1

    RSN-07 reinforcement learning with rule-based rewards

17. S17 Ouyang et al., InstructGPT

    PST-01 instruction tuning and human feedback

18. S18 Zhou et al., LIMA

    PST-02 small curated SFT datasets

19. S19 Singhal et al., length correlations in RLHF

    PST-04 length bias

20. S20 Christiano et al., RL from human preferences

    PST-05 learned rewards from comparisons

21. S21 Stiennon et al., learning to summarize from human feedback

    PST-05 RLHF for language

22. S22 Gao et al., reward model overoptimization

    PST-05 proxy rewards and the KL penalty

23. S23 Rafailov et al., DPO

    PST-06 loss formula used in the lab

24. S24 Hu et al., LoRA

    PST-07 parameter counts

25. S25 Dettmers et al., QLoRA

    PST-07 4-bit fine-tuning

26. S26 Hinton et al., distillation

    PST-08 student and teacher models

27. S27 Qi et al., fine-tuning compromises safety

    PST-08 safety regressions

28. S28 Amodei et al., concrete problems in AI safety

    SAF-01 technical risk categories

29. S29 Askell et al., a general language assistant

    SAF-02 helpful, honest, harmless

30. S30 Bai et al., helpful and harmless assistant

    SAF-02 RLHF for harmlessness

31. S31 Bai et al., Constitutional AI

    SAF-02 principle-based self-critique

32. S32 Wei et al., Jailbroken

    SAF-02 why safety training fails

33. S33 Zou et al., universal adversarial attacks

    SAF-02 transferable adversarial suffixes

34. S34 Greshake et al., indirect prompt injection

    SAF-04 attacks through retrieved content

35. S35 Perez et al., red teaming with language models

    SAF-07 automated red-teaming

36. S36 Ganguli et al., red teaming to reduce harms

    SAF-07 human red-team dataset

37. S37 OWASP Top 10 for LLM Applications

    SAF-03 application risk categories