12 free graduate ML textbooks — curation source for the RDCO bookshelf

Why this is in the vault

Founder shared the article on 2026-04-30 with the framing “I think our setup is more advanced” — meaning the article’s “upload PDFs into a Claude Project as ambient context” workflow is naive vs. the bookshelf architecture proposed in 2026-04-30-bookshelf-source-material-architecture-gap. But the curation list itself is genuinely valuable as a starter set for the ML-fundamentals domain when bookshelf scaffold lands. Filing for that.

The 12 textbooks (with corrected attribution)

The article’s “MIT graduate-level AI textbooks” framing is loose marketing — only ~2-3 of the 12 are actually MIT. The real story: these are the canonical free graduate ML textbooks across the field, regardless of host institution.

Foundations

1. Foundations of Machine Learning — Mohri, Rostamizadeh, Talwalkar
   • Institution: NYU
   • URL: http://mlbook.cs.nyu.edu
   • What it covers: theoretical math behind why learning algorithms work
2. Understanding Deep Learning — Simon J.D. Prince
   • Institution: University of Bath
   • URL: https://udlbook.github.io/udlbook
   • What it covers: clearest in-print explanation of how neural nets process and weight information
3. Machine Learning Systems — Vijay Janapa Reddi (lead) + community
   • Institution: Harvard / MIT (textbook is part of the broader MLSysBook collaboration; “MIT curriculum” framing in the article is closest to accurate here)
   • URL: https://mlsysbook.ai
   • What it covers: bridges theory to production; how models are built, deployed, and fail in real environments

Advanced techniques

4. Algorithms for Decision Making — Kochenderfer, Wheeler, Wray
   • Institution: Stanford
   • URL: https://algorithmsbook.com
   • What it covers: sequential decisions as policy problems under uncertainty
5. Deep Learning — Goodfellow, Bengio, Courville
   • Publisher: MIT Press (this is the only legitimate “MIT” thread in the article’s framing)
   • URL: https://www.deeplearningbook.org
   • What it covers: the original deep learning textbook; dense but foundational

Reinforcement learning

6. Reinforcement Learning: An Introduction — Sutton & Barto
   • Institution: University of Alberta + DeepMind
   • URL: http://incompleteideas.net/book/the-book.html
   • What it covers: the canonical RL textbook; every agent that reasons sequentially traces back here
7. Distributional Reinforcement Learning — Bellemare, Dabney, Rowland
   • Institution: DeepMind / Mila / Google Brain
   • URL: https://www.distributional-rl.org
   • What it covers: shifts the frame from expected rewards to full probability distributions over outcomes
8. Multi-Agent Reinforcement Learning — Albrecht, Christianos, Schäfer
   • Institution: University of Edinburgh
   • URL: https://marl-book.com
   • What it covers: how multiple decision-making agents learn to interact in shared environments
9. Decision Making Under Uncertainty (long-game) — Kochenderfer
   • Institution: Stanford
   • URL: https://mykel.kochenderfer.com/textbooks
   • What it covers: probabilistic frameworks for decision-making across long time horizons

Ethics + probability

10. Fairness and Machine Learning — Barocas, Hardt, Narayanan
    • Institutions: Microsoft Research / UC Berkeley / Princeton
    • URL: https://fairmlbook.org
    • What it covers: where models break down and where bias enters; required reading for knowing when NOT to trust model output
11. Probabilistic Machine Learning: An Introduction — Kevin Murphy
    • Institution: Google Research
    • URL: https://probml.github.io/book1.html
    • What it covers: builds every concept on probability theory from the ground up
12. Probabilistic Machine Learning: Advanced Topics — Kevin Murphy
    • Institution: Google Research
    • URL: https://probml.github.io/book2.html
    • What it covers: causal inference, decision making under uncertainty, generative models at graduate level

Mapping against Ray Data Co

As a bookshelf starter set

When the bookshelf scaffold (2026-04-30-bookshelf-source-material-architecture-gap) lands, this list is a clean candidate for the ML-fundamentals domain. Recommended starter triad (per the article’s own claim, which is defensible):

• Murphy Vol I (Probabilistic ML) — single most-cited foundation
• Mohri (Foundations of ML) — formal theory backbone
• Sutton & Barto — RL canon, directly relevant to agent architecture work

Adding the full 12 takes ~5-10GB of disk and one afternoon of ingestion work. Defer until founder greenlights the bookshelf scaffold.

Why our (proposed) system is structurally more advanced than the article’s

Both work. Dami’s is right for “I want to test this in an afternoon.” Ours is right for “I’m running an agent-native company and need this as durable infrastructure.”

Why filing this curation matters even before bookshelf-build

Even if the bookshelf scaffold takes another month to land, this curation note is useful as:

• A reference for the founder’s own reading queue (he could grab any one of the 12 and start reading without waiting for Ray’s retrieval layer)
• A starter set for the future “domain backfill” decision (when bookshelf-build green-lights, this is the ML-fundamentals canonical list)
• A vocabulary anchor — when Ray says “Wheeler argues X” in the future, the citation chain points HERE for the specific Wheeler book (Algorithms for Decision Making at Stanford, NOT a generic memory)

Caveats on the article’s framing

The article (Dami-Defi, X) is heavily marketing-flavored:

• “MIT curriculum” framing is misleading (only ~2-3 are actually MIT)
• “What happened next is why I don’t use the same research setup I had 30 days ago” — typical engagement-bait opening
• The Kelly Criterion anecdote is real and useful (Murphy’s Vol II does cover heavy-tailed inference; the math is correct), but the dramatic framing inflates a routine retrieval result
• The author is in the crypto/marketing space (per their X bio: “@bitget Partner”), not academic ML — they’re popularizing, not authoring the field
• The “right now most people on CT are using Claude as a search engine” framing is half-true and half-FOMO

What’s actually load-bearing in the article:

• The 12-book list (filed above)
• The point that ambient-context-PDFs upgrades reasoning quality (true, well-documented)
• The “afternoon to set up” claim (true for the Dami pattern; less true for the bookshelf pattern but bookshelf has compounding value)

Related

• 2026-04-30-bookshelf-source-material-architecture-gap — the proposed bookshelf architecture this curation feeds into
• 2026-04-30-quality-gate-as-brain-org-boundaries-agentic-companies — bookshelf is the input layer the quality gate evaluates against
• 2026-04-30-mitohealth-founder-5-layer-agent-native-company-loop — Layer 1 (sensors+data) home for the bookshelf
• Notion Research Backlog: queue “ML-fundamentals bookshelf backfill — 12 textbook starter set” as decision item if founder green-lights the scaffold