390 saved links, 2017–2026 (peak: 2023). The theory shelf of your AI transition: canonical courses and free books, an unusually deep reinforcement-learning thread that suddenly became career-relevant when RLHF and agent-training arrived, and a running subscription to deeplearning.ai’s short-course machine. Less newsy than your other AI tags — most of what’s here stays true.
Related: Machine Learning · LLMs · NLP · Data Science · Python
The canon — courses
- MIT 6.S191 — Introduction to Deep Learning — refreshed yearly, saved yearly.
- fast.ai’s Practical Deep Learning for Coders — the top-down classic.
- Karpathy’s Neural Networks: Zero to Hero — and its fan tributes: nanoGPT and a GPT from scratch in Go, fully commented.
- Stanford CS230 · CMU 11-785 · Berkeley CS182 — the lecture-video shelf.
- Learn PyTorch in a day — literally (video).
The canon — books
- Dive into Deep Learning — interactive, code + math + visualizations; its math appendix is the best math-for-DL resource you’ve saved.
- Understanding Deep Learning (Simon Prince) — the free 541-page PDF.
- deeplearningbook.org — Goodfellow, Bengio, Courville; the original.
- Deep Learning with Python — Chollet’s step-by-step notebooks (“highest quality I’ve seen anywhere”) and the book’s free web edition.
- The Little Book of Deep Learning · Grokking Deep Learning — the compact companions.
Transformers & attention, visually
- 3Blue1Brown: Transformers and attention, step by step — the visual intuition, chapters 5 and 6.
- implementing “Attention Is All You Need” from scratch · transformers by hand ✍ · the well-explained Transformer guide with Keras code.
- building Transformers from first principles — course lecture + PyTorch notebooks.
- Vision Transformers, explained · CNNs and vision architectures course.
- CNN Explainer — interactive visualization for non-experts.
- Shreya Rao’s “Deep Learning Illustrated” series — how a network learns, patiently drawn.
Reinforcement learning — the thread that came back
- Nathan Lambert’s RLHF book — “one of the best resources”; where classic RL meets modern LLMs.
- David Silver’s RL course and Stanford CS234 — the foundations.
- the mathematical foundation of RL book — “in the beginning, ML was RL.”
- the hands-on RL course · Hugging Face’s deep RL course · 10 repos to master RL.
- Pearl — Meta’s production-ready RL agent library.
- The agent era: Experiential Reinforcement Learning — agents that learn from experience · infrastructure for agent RL · LlamaGym — fine-tune LLM agents with online RL.
The deeplearning.ai subscription
Andrew Ng’s short-course machine — the tag’s steady drip.
- the Attention course · Pretraining LLMs · Finetuning LLMs · Improving Accuracy of LLM Applications.
- Agent-flavored: long-term agentic memory (from the MemGPT authors) · Practical Multi AI Agents with crewAI · Anthropic’s Agent Skills course.
- The Batch — the newsletter thread running through the whole tag.
Practice & production
- Stas Bekman’s Art of Debugging — the ML debugging open book.
- slaying OOMs with FSDP and torchao (video) · Unsloth: fine-tune and deploy LLMs on your phone.
- the Flow Matching guide with PyTorch code — “great introduction, fantastic visualizations.”
- deep learning in production — deployment notes and references.
- from zero to training state-of-the-art OSS models in a year — proof the ramp is climbable.