Books I want to read

• Reinforcement Learning from Human Feedback — Manning (via AI Book Club)
• AI Product Manager’s Handbook (2nd ed.) — Irene Bratsis, Packt (building/scaling AI products, AI-native vs. evolving products, commercialization)
• Machine Learning Platform Engineering — Benjamin Tan Wei Hao, Shanoop Padmanabhan, Varun Mallya, Manning (build an MLOps/LLMOps platform from scratch: Kubeflow, MLflow, BentoML, Feast, model serving & monitoring)
• Machine Learning System Design — Valerii Babushkin, Arseny Kravchenko, Manning (end-to-end ML system framework: problem framing, dataset gathering, training pipelines, serving & monitoring)
• Knowledge Graphs and LLMs in Action — Giuseppe Futia, Vlastimil Kus, Manning (building knowledge graphs from structured/unstructured sources, integrating with LLM apps & RAG pipelines)
• Graph Neural Networks in Action — Keita Broadwater, Manning (building GNNs in Python for node prediction, link prediction, graph classification)
• Introduction to Machine Learning — Laurent Younes, arXiv textbook (mathematical foundations of ML: linear algebra/probability, kernel methods, supervised & generative learning, generalization theory)
• The Little Book of Deep Learning — François Fleuret (compact free ebook: gradient descent, backprop, model components, architectures, applications)
• Mathematical Introduction to Deep Learning: Methods, Implementations, and Theory — Kuckuck et al., arXiv (737pp: ANN architectures, optimization theory, approximation/generalization theory, deep learning for PDEs)
• LLM Customization and Fine-Tuning — Amit Bahree, Weehyong Tok, Manning (adaptation spectrum from prompting/RAG through LoRA/QLoRA, full SFT, distillation, and DPO alignment; production ops for drift & safety)
• A First Course in Causal Inference — Peng Ding, UC Berkeley lecture notes (causal inference from basic probability, statistical inference, linear/logistic regression)
• AI Agents and Applications: With LangChain, LangGraph, and MCP — Roberto Infante, Manning (build LLM-powered agentic applications: agent workflows, tools, MCP integrations)
• Multi-Agent AI Engineering: Design, build, and operate AI systems that think and act as coordinated teams — Dr. Xiao Ma, Dr. Chi Wang, Packt (production-grade multi-agent systems: communication protocols, memory/context, orchestration, evaluation, security, observability)
• AI Agents in Action (2nd ed.) — Micheal Lanham, Manning (autonomous agent design/deployment, MCP tools/memory, reasoning & planning patterns — ReAct, Reflexion, Tree-of-Thought, multi-agent patterns)

AI Plays Games: From Atari to Agents

A study track on how machines learn to play video games, from reinforcement learning on raw pixels to LLM agents that write their own skills.

Foundations

• Reinforcement Learning: An Introduction (2nd ed.) — Richard Sutton, Andrew Barto, MIT Press, 2018 (the standard RL textbook; the authors offer a free PDF, linked here via a Stanford course copy)

Reinforcement learning from pixels

• Playing Atari with Deep Reinforcement Learning — Mnih et al., DeepMind, 2013 (the starting point: one network learns Atari games from raw pixels)
• Human-level control through deep reinforcement learning — Mnih et al., Nature, 2015 (the DQN paper)
• Mastering the game of Go with deep neural networks and tree search — Silver et al., Nature, 2016 (AlphaGo)
• Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model — Schrittwieser et al., 2019 (MuZero: plans with a model it learns itself, no rules given)

Large-scale competitive play

• Grandmaster level in StarCraft II using multi-agent reinforcement learning — Vinyals et al., Nature, 2019 (AlphaStar)
• Dota 2 with Large Scale Deep Reinforcement Learning — OpenAI, 2019 (OpenAI Five)

LLM agents and social games

• Voyager: An Open-Ended Embodied Agent with Large Language Models — Wang et al., 2023 (Minecraft agent that builds a reusable skill library as code)
• Human-level play in the game of Diplomacy by combining language models with strategic reasoning — Meta FAIR Diplomacy Team, Science, 2022 (Cicero: negotiates in natural language)

Articles

• Extending Raschka’s GPT-2: an MoE trained from scratch on an RTX 3090 — Giles Thomas
• AI Infra: 大模型系统设计与工程实践 — 李博杰 (open-source book on AI infrastructure: model serving, distributed training/inference, hardware & compute estimation)
• James H. Simons, PhD: Using Mathematics to Make Money — James Simons interview, Journal of Investment Consulting (quant investing at Renaissance Technologies: model-building, hiring scientists over finance veterans, collaboration)

Blogs

• Brendan Gregg’s Homepage — systems performance engineer (creator of Flame Graphs, eBPF/BPF tools, DTrace; author of Systems Performance and BPF Performance Tools); site collects his docs, talks, and tools on Linux/cloud performance analysis
• 科学空间 (Scientific Spaces) — 苏剑林, blog on math/ML theory (optimizer theory — Adam/Muon, scaling laws, manifold optimization, information theory)
• Sebastian Raschka, PhD — LLM research engineer, author of Build a Large Language Model (From Scratch); blog covers LLM research, architecture notes (Kimi K3, Muse Glimmer), and practical/code-driven AI deep dives
• Paul Graham’s Essays — startups, tech, and how to think, from the Y Combinator co-founder
• RLHF Book — Nathan Lambert, living online textbook on Reinforcement Learning from Human Feedback (RLHF fundamentals, reward modeling, PPO/DPO, RLHF’s role in modern LLM post-training)
• Lil’Log (Lilian Weng) — long survey-style posts (20 to 40 minute reads); latest: Harness Engineering for Self-Improvement (2026.7), Scaling Laws, Carefully (2026.6)
• Andrej Karpathy blog — infrequent posts; latest: microgpt (2026.2), GPT training and inference in 200 lines of pure Python
• Tri Dao’s blog — FlashAttention author; 2026 posts include Gram Newton-Schulz (fast Newton-Schulz for Muon), SonicMoE, ReplaySSM
• Giles’ Blog (Giles Thomas) — tutorial-style posts on learning LLMs by building them, “the post I wished I’d found when I started learning”; includes a series working through Raschka’s from-scratch book; actively updated (2026.10)
• Interconnects — Nathan Lambert, newsletter on open models, post-training and the AI industry
• Tim Dettmers — quantization, GPU hardware and open coding agents (Building SERA, 2026.1)
• Simon Willison’s Weblog — daily notes on LLM tools, coding agents and security
• Hamel Husain’s Blog — applied AI engineering, focused on evals
• Eugene Yan — applied LLM systems and recommendation systems
• Sasha Rush — post-training researcher at Cursor; hands-on exercises: Thinking like Transformer, LLM Training Puzzles
• Thonk From First Principles (Horace He) — ML systems from first principles
• Transformer Circuits — interpretability research thread; Chris Olah’s current writing, moved from his older blog colah.github.io (last updated 2021)
• The Scaling Hypothesis (Gwern) — long essay, not a running blog
• LessWrong — community blog on rationality and AI alignment; user-submitted essays, with the Sequences (Eliezer Yudkowsky’s rationality essays) as its core reading
• Facing the Intelligence Explosion (Luke Muehlhauser) — a short online book on what happens when machines become better than humans at doing science and engineering
• Recommended Readings (Luke Muehlhauser) — a deliberately short reading list, at most three sources per topic, covering effective altruism, existential risk, extreme-stakes AI risk, forecasting and more

Code Repositories

• Axolotl — open-source LLM fine-tuning framework (LoRA/QLoRA, distributed training with FSDP2 and DeepSpeed; Apache-2.0)
• LitGPT — Lightning AI, 20+ high-performance LLMs with recipes to pretrain, finetune and deploy at scale (Apache-2.0)

Movies

Big ideas

• Arrival — Denis Villeneuve, 2016 (a linguist communicates with aliens; language, time and family)
• Contact — Robert Zemeckis, 1997 (science and faith)
• Interstellar — Christopher Nolan, 2014 (space travel and time)
• 2001: A Space Odyssey — Stanley Kubrick, 1968 (the founding work of the genre; very slow)
• The Martian — Ridley Scott, 2015 (survival by hard science, light and optimistic)

Solitude, vision, reflection

• Oblivion — Joseph Kosinski, 2013 (a lone drone repairman on a post-apocalyptic Earth)
• Blade Runner 2049 — Denis Villeneuve, 2017 (identity and memory)
• Moon — Duncan Jones, 2009 (low budget, one man on a lunar base; identity and clones)
• Ex Machina — Alex Garland, 2014 (artificial intelligence and what makes us human)
• Annihilation — Alex Garland, 2018 (a mutating zone with strange rules; needs patience)
• Gravity — Alfonso Cuarón, 2013 (survival in space)
• Ad Astra — James Gray, 2019 (father and son, inward-looking)

Time and causality

• Tenet — Christopher Nolan, 2020 (time inversion; worth planning to watch twice)
• Primer — Shane Carruth, 2004 (ultra-low budget, a rigorously worked-out time machine)
• 12 Monkeys — Terry Gilliam, 1995 (a time loop and fatalism)
• Coherence — James Ward Byrkit, 2013 (low budget, parallel worlds)

Reality and perception

• Inception — Christopher Nolan, 2010 (layered dreams with clear rules)
• The Matrix — the Wachowskis, 1999 (the classic of virtual reality)
• The Prestige — Christopher Nolan, 2006 (not pure science fiction; intricate structure)
• Shutter Island — Martin Scorsese, 2010 (psychological mystery with a strong twist)

Closed spaces

• Cube — Vincenzo Natali, 1997 (strangers trapped in a cube of traps)