Reading paths
Series
Start with the problem you want to solve. You do not need to remember the series name.
112 published series
AI Agent Arxiv Digest
Posts in the AI Agent Arxiv Digest series
AI Daily
A daily digest of AI developments.
AI Agent GitHub Digest
Posts in the AI Agent GitHub Digest series
AI Engineer Interview Daily
A daily AI engineer interview drill rotating through seven topics by day of the week — ML fundamentals, deep learning, system design, LLM engineering, coding, paper reading, and behavioral — pulling the latest interview questions and resources from the web.
AI Framework Changelog
Posts in the AI Framework Changelog series
AI Agent Funding
Posts in the AI Agent Funding series
AI Model Tracker
Posts in the AI Model Tracker series
AI Pricing Watch
Posts in the AI Pricing Watch series
Product Builder Interview Drill
Posts in the Product Builder Interview Drill series
AI Region Focus
Posts in the AI Region Focus series
AI Tool of the Day
Posts in the AI Tool of the Day series
AI Agent Weekly Review
A weekly roundup of notable releases, papers, and tooling changes in the AI agent space.
Product Builder Interview Daily
A daily product builder interview drill rotating through seven topics by day of the week — product sense, metrics, strategy, AI product design, growth, technical PM, and behavioral — pulling the latest case studies and interview questions from the web.
AI Security Alert
Posts in the AI Security Alert series
Reading CMU 10-423
Reading CMU 10-423/623/723 Generative AI through the Spring 2026 edition: 26 lecture decks, HW1–HW4 starter code, and the practice exam, from language models to diffusion, multimodal models and scaling.
Reading CMU 11-868 LLM Systems
A lecture-by-lecture reading of CMU 11-868 LLM Systems (Spring 2026) through its 28 public slide decks and seven MiniTorch assignments, from CUDA kernels and a homemade framework to distributed training, serving, and RLHF, with the no-video, GPU-required limits for self-learners noted throughout.
Reading Stanford CS149
A lecture-by-lecture reading of Stanford CS149 Parallel Computing (Fall 2025) using its official slides, five programming assignments, and four written assignments: multi-core and SIMD, work distribution and locality, GPUs and CUDA, DNNs and AI accelerators (Trainium2), datacenter AI, AI-driven optimization, then cache coherence, lock-free programming, and transactional memory, with the public 2023 videos as a labeled supplement.
Reading Stanford CS224R
A reading of Stanford CS224R Spring 2026 through its 17 slide decks, three homeworks, and default project. It covers deep RL from imitation learning, policy gradients, and offline RL to RLHF, LLM reasoning, and robot VLAs, with the public Spring 2025 videos as a labeled supplement.
Reading Stanford CS231N
A guided reading of Stanford CS231N (Deep Learning for Computer Vision), based on the Spring 2026 slides and assignments A1–A3, with the public Spring 2025 YouTube lectures for video. It runs from image classification, backprop, CNNs and Transformers through detection and segmentation, self-supervised learning, generative models, vision-language and 3D.
Reading Stanford CS234
Reading Stanford CS234 Reinforcement Learning through the Winter 2026 slides for 14 lectures and three assignments with starter code, alongside the public Spring 2024 videos: MDP planning, model-free evaluation and control, policy gradients and PPO, imitation learning and RLHF/DPO, exploration theory with bandits, MCTS, and value alignment.
Reading Harvard CS2881R
A lecture-by-lecture reading of Harvard CS 2881R AI Safety (Boaz Barak, Fall 2025). It starts from the emergent-misalignment HW0, then covers safety training, jailbreaks and prompt injection, model specs and content policies, scheming and interpretability, recursive self-improvement, capability measurement, and the economic and mental-health impacts, ending with the students' reproduction and final research projects. It is based on the public recordings, reading lists, slides and assignment specs.
Reading MIT 6.5940
A lecture-by-lecture reading of MIT 6.5940 TinyML and Efficient Deep Learning Computing, based on the latest complete edition (Fall 2024): pruning, quantization, NAS, distillation, microcontroller deployment, LLM inference and post-training, long context, diffusion, and distributed and on-device training, cross-referenced with the in-progress Fall 2026 offering.
Reading MIT 6.S184
A lecture-by-lecture reading of MIT 6.S184 (IAP 2026) from the official lecture notes, slides, recordings, and three labs with solutions: ODEs/SDEs, flow matching, score matching, classifier-free guidance, DiT and latent spaces, and discrete diffusion.
Reading NCCU Yen-Lung Tsai Generative AI
A lecture-by-lecture reading of NCCU Yen-Lung Tsai's TAICA course "Generative AI: Text and Image Synthesis Principles and Practice", based on the Spring 2025 (1132) term: 14 recordings, 14 slide decks, 12 homework specs, and Colab notebooks. It runs from neural nets, GANs, LLMs and Transformers through chatbots, RAG and AI agents to VAEs, Stable Diffusion, and ControlNet/Fooocus, and is written for beginners.
Reading NTHU Hung-Yu Kao Natural Language Processing
Reading NTHU Prof. Hung-Yu Kao's Mandarin TAICA NLP course through its complete Fall 2025 materials on the official IKMLab GitHub: lecture slides, 36 public W1–W16 recordings, HW1–HW4 specs and notebooks, and TA tutorials on PyTorch, Hugging Face, LLM APIs and RAG. The series moves from classic text processing, word embeddings, seq2seq, Transformers and the BERT family through decoding and evaluation to RLHF, PEFT, RAG and reasoning, and ends with what changes in Fall 2026.
Reading NTU Yun-Nung Chen Applied Deep Learning 2025 Fall
Reading NTU Yun-Nung (Vivian) Chen's Applied Deep Learning (ADL) Fall 2025 through its 17 lecture decks, 77-video playlist, TA recitations and the HW1 spec: neural nets, RNNs, Transformers, BERT, pretraining, RLHF, LoRA, RAG, decoding, safety and alignment, language agents, and reasoning.
Reading NTU Hsuan-Tien Lin Machine Learning Foundations & Techniques
A topic-by-topic guide to Hsuan-Tien Lin's Machine Learning Foundations and Techniques MOOCs (32 lectures, 130 YouTube videos, all handout slides). It runs from PLA, VC dimension, linear models, regularization, and validation to SVMs, kernels, aggregation, tree models, and neural networks. The public Fall 2024 HW0–HW7 and final project serve as exercises, and the in-progress Fall 2026 offering is cross-referenced.
Reading NTU Hung-yi Lee Machine Learning 2026 Spring
Reading NTU Hung-yi Lee's Machine Learning 2026 Spring through its 8 lecture decks and recordings plus 10 homework Colabs: an OpenClaw teardown, context engineering, Flash Attention, KV cache, positional embedding, harness engineering, self-correction, and self-improving AI.
Global AI/CS Course Map
A 2025–2026 guide to AI and CS course access at Stanford, CMU, MIT, and UC Berkeley, distinguishing complete self-study courses from public syllabi, partial materials, and historical videos.
Reading Stanford CME295
A lecture-by-lecture reading of Stanford CME295: Transformers & Large Language Models, from the Transformer architecture to LLM evaluation and AI agents, and how it divides the ground with CS224N and CS336.
Reading CMU 11-768 AI Agents
A lecture-by-lecture reading of CMU 11-768 AI Agents (Fall 2026), from the agent loop and tool use to RL training, credit assignment, and reward hacking, following the official course order.
Reading Stanford CS224U
A unit-by-unit reading of a versioned Stanford CS224U offering: semantic representations, natural-language inference, question answering, and interactive language systems.
Reading Berkeley CS189
A lecture-by-lecture, homework-by-homework reading of the public materials of Berkeley CS189 (Introduction to Machine Learning), with the term stated in every post, filling in the fuller mathematical foundations of ML after CS188.
Reading CMU 07-380
Continuing from 07-280, a lecture-by-lecture reading of CMU 07-380’s first-offering 26 lectures — from logic and planning through diffusion models — with the release status of homework and project materials noted throughout.
Harvard CS181 Weekly Guides
A week-by-week reading of Harvard CS181 (Machine Learning): the linear algebra, calculus, and probability refreshers, then linear regression and the models that follow, based on public homework.
AI Agent Memory Engineering
Why memory is the core hard problem of agent engineering: ten posts, each tackling one memory problem, from a full context window to choosing an open-source memory framework.
Deep Research Frontier
A survey of 80+ Deep Research systems: the three-stage capability roadmap, four core components (planning, acquisition, memory, generation), three optimization paradigms (prompting, SFT, RL), evaluation, and the open-source tool landscape.
AI Model Families
Tracing the evolution, architecture, licensing traps, and version selection of mainstream model families — Qwen, DeepSeek, Claude, GPT, Gemini, Llama, Mistral, GLM, Kimi — with pick guidance for agent developers.
Multi-Agent Systems in Practice
Comparing the subagent models, orchestration patterns, and communication mechanisms of Claude Code, Codex, Antigravity, Cursor, Windsurf, Devin, LangGraph, and CrewAI, with a capability matrix and a spectrum of design philosophies.
How Free Content Acquires Customers for Another Business
Posts in the How Free Content Acquires Customers for Another Business series
AI Search Is Rewriting the Content Business
How AI summaries redraw the path from content to citation, click, and conversion, drawing on data from Google, Pew, and Cloudflare, and which part of the content business blocking, licensing, lawsuits, and owned assets each protect.
Who Controls the Creator-Reader Relationship
Posts in the Who Controls the Creator-Reader Relationship series
How Intelligence Becomes an Enterprise Business
Posts in the How Intelligence Becomes an Enterprise Business series
Document Parsing in Practice
The three-layer ladder for turning documents into LLM-readable content — conversion, extraction, and parsing. From picking the right layer to comparing MarkItDown, anydoc, MinerU, and the rest.
Content Selling Business Models
Posts in the Content Selling Business Models series
AI-Native SDLC Playbook
A playbook for putting AI into every stage of the software development lifecycle: how agents fit into requirements, design, development, testing, and operations.
Groundlane 實戰系列
Notes on designing and using Groundlane, a self-hosted research and scraping tool — a companion case study to the search-and-scraping series.
Reading Stanford CS329Z
A lecture-by-lecture reading of Stanford CS329Z on agent engineering, written only as current official materials appear rather than treating a tentative syllabus as delivered instruction.
Reading Stanford's Main-Line CS Courses
A map of Stanford CS core courses, from the degree foundations through AI, NLP, graph learning, and agents, with versioned course guides and prerequisites.
資安證照攻略
Preparation paths for security certifications built on official exam guides: what each domain tests, which official material covers it, and what to practice.
從零訓練一個 LLM
A hands-on record of training a language model from scratch: every decision across data, tokenizer, architecture, training, and evaluation.
Learning Coding-Agent Design from Mature Systems
A Looplane-driven comparison of pi, OMP, OpenCode, Codex CLI, and Claude Code, from loops, workspaces, approvals, and verification through shipped baselines for memory, compaction, MCP, sandboxing, subagents, replay, LSP, cost tracking, and Agent as a Service, with production validation and runtime-parity gaps kept explicit.
The RAG Techniques Compendium
RAG taken apart into techniques you can compare one at a time: chunking and indexing, sparse and dense retrieval, ranking and fusion, agentic and advanced patterns, generation-side control, the failure modes real queries hit, and evaluation, cost and observability. One decision per post, assembled into a pipeline of your own.
AI Benchmark Watch
Posts in the AI Benchmark Watch series
OMP Internals Deep Dive
A close reading of oh-my-pi (OMP) source: streaming internals, the rulebook and TTSR, slash and custom tools, memory, and the task hub/advisor design.
pi-mono Deep Dive
A close reading of the pi-mono source: how it implements the agent loop, tool calls, approvals, and session management — one of the reference implementations in the mature-coding-agent-design comparison.
Ask AI in Practice
Follow the real quidproquo Ask AI data path from indexing, hybrid retrieval, writing, and source gates through streaming, caching, incident analysis, and reproducible evaluation. Each post traces one responsibility and the boundary of what its evidence can prove.
Cloudflare AI Stack
The AI-specific pieces of Cloudflare’s platform, read on their own: Workers AI, Vectorize, AI Gateway, and the trade-offs of building on these bindings.
Cloudflare Edge Platform
An ongoing record of updates to Cloudflare’s edge platform components and services, extending past where "The Cloudflare Edge Stack" series left off.
Looplane Architecture Notes
Follow one coding-agent task through Looplane: from the TUI, disposable workspace, prompt, and two runtime lanes through tool authority, the state/event lifecycle, MCP, subagents, SDK/IDE integrations, and finally Cloudflare remote execution. Each article traces one data flow, failure boundary, and test surface.
Reading Harvard CS50 AI
A lecture-by-lecture reading of Harvard CS50’s AI with Python: search, knowledge representation, probability, machine learning, neural networks, and language, based on the public course materials and assignments.
MIT 6.7960 導讀 (Fall 2024 OCW)
A lecture-by-lecture reading of MIT 6.7960 (Fall 2024 OCW): optimization, regularization, CNNs, Transformers, generative models, and representation learning in deep learning.
Search and Scraping in Practice
The full path for getting data in from outside: renting a cloud search API versus self-hosting one, choosing among the scraping tools, what to do when anti-bot defenses block you, and how to wire it all into a research pipeline. One decision per post.
Statistics from Exams to ML/AI
A statistics learning path that starts from NTU IM exam preparation, builds through statistical inference and applied modeling, and connects each topic to ML/AI training, evaluation, experiments, and data workflows.
Meta-Harness 與 Agent 治理
When the agent itself needs governing: meta-harness design, and the rules and boundaries for multi-agent collaboration.
認識 AI 模型
An introductory series for readers still new to AI models: concepts, terminology, and how to pick one to use.
一個人的媒體公司
A practical record of running a one-person media business: positioning, output cadence, monetization, and where AI tools fit into the workflow.
Claude Code Deep Dives
Taking apart Claude Code itself — CLI architecture, the permission model, and internal mechanics — filling in the product-internals detail that the automation guide series does not cover.
Self-Hosted Inference
The trade-offs of running your own inference service: hardware, model-serving frameworks, cost, and operations, weighed against just calling a cloud API.
Reading AI Top Conferences
How AI top conferences are recognized, how submissions and review work, and how the flagship venues differ.
AI Conference Guide
Posts in the AI Conference Guide series
Choosing an Agent CLI
A comparison of terminal agents — Claude Code, Codex, Gemini CLI (now transitioned to Antigravity CLI), OpenCode, Pi, Cursor CLI, and Kiro — covering each one's design trade-offs, plans, and billing, closing with a cross-tool subscription comparison and multi-model routing. Pricing and model names rot fast, so every post carries its verification date and defers the perishable details to official pages.
Technology Choices in the AI Era
Adoption remains the primary criterion, augmented by five AI-era criteria — machine-readable docs, types, source-in-repo, data skeleton, and machine-callability — from frontend to backend, cloud to self-hosted.
Reading CMU 07-280
A lecture-by-lecture reading of CMU 07-280, the first half of CMU’s redesigned AI core: linear regression, MLE, N-grams, attention/transformers, and Q-learning, against the official Spring 2026 materials.
Reading CMU 11-785 Deep Learning
A lecture-by-lecture reading of CMU 11-785 Spring 2026 that separates its public 28-lecture teaching sequence from the restricted assignment workflow.
Reading Stanford CS124
A week-by-week reading of Stanford CS124: language models, text classification, information extraction, question answering, speech, and the full NLP pipeline.
Reading Stanford CS224N
A lecture-by-lecture reading of Stanford CS224N: word vectors, sequence models, Transformers, large language models, evaluation, and responsible NLP.
Reading Stanford CS224V
A unit-by-unit reading of one explicitly versioned Stanford CS224V offering: understanding, dialogue management, generation, evaluation, and deployment for conversational assistants.
Reading Stanford CS336
A lecture-by-lecture reading of Stanford CS336: tokenizers, data, scaling, training, parallelism, evaluation, and alignment across the full language-model pipeline.
Reading MIT 6.S191
Reading all nine lectures and three labs of MIT 6.S191 from the official 2026 videos, slides, and lab code without mixing in earlier offerings.
Private Corpus Pipeline
How private data enters indexes safely and continuously, remains subject to query-time authorization, and stays consistent when sources change or disappear—focused on the data lifecycle rather than RAG retrieval techniques.
Reading Stanford CS221
A lecture-by-lecture reading of Stanford CS221: search, Markov decision processes, machine learning, constraint satisfaction, and probabilistic models.
Reading Stanford CS224W
A lecture-by-lecture reading of Stanford CS224W: graph representation, network science, graph neural networks, knowledge graphs, and scalable graph learning.
Reading Stanford CS229
A chapter-by-chapter reading of Stanford CS229’s official 2026 notes, spanning supervised and deep learning, foundation models, LLM reasoning, and reinforcement learning across twenty-one chapters without pretending to reconstruct a single quarter’s lecture schedule.
Berkeley CS188 Spring 2026
Reading Berkeley CS188 Spring 2026 through Projects P0–P5, from search and decision making to probabilistic inference, reinforcement learning, and machine learning.
Reading Berkeley CS285 Spring 2026
Reading Berkeley CS285 Spring 2026 in deep reinforcement learning through 25 lectures, nine discussions, five assignments, and their compute constraints.
Berkeley CS288 Spring 2026
Reading Berkeley CS288 Spring 2026 from n-grams through RAG, reasoning, and agents using its 18 public slide units and three assignments.
Reading CMU 10-301 Machine Learning
Reading the 27 lectures of CMU 10-301/601 through its nine public Spring 2026 homework bundles and the practical limits for independent learners.
Reading Stanford CS107
A lecture-by-lecture reading of Stanford CS107: C, memory, assembly, data representation, and systems debugging from high-level code down to the machine.
Reading Stanford CS103
A lecture-by-lecture reading of Stanford CS103: discrete mathematics, logic, proofs, sets, computability, and the shared language they provide for later CS courses.
Reading Stanford CS109
A lecture-by-lecture reading of Stanford CS109: probability, random variables, inference, and simulation as the foundation used by machine learning and data science.
Reading Stanford CS111
A lecture-by-lecture reading of Stanford CS111: processes, threads, synchronization, virtual memory, file systems, and operating-system design trade-offs.
Reading Stanford CS228
A week-by-week reading of one explicitly versioned Stanford CS228 offering: probabilistic graphical models, exact and approximate inference, and parameter and structure learning.
Reading Stanford CS161
A lecture-by-lecture reading of Stanford CS161, Winter 2026: algorithm design, correctness proofs, and complexity analysis across all eighteen public lecture units.
AEO, GEO, and AI Search
Writing for a reader that is now a model: from the SEO groundwork through answer engine optimization, what content structure and structured data actually buy, and whether the tracking tools can really measure visibility inside AI search.
AI Engineer Interview Prep
Preparing for AI engineer interviews across ten topics — ML fundamentals, system design, LLM application architecture, coding, paper reading, and behavioral. Each post focuses on one interview dimension with core concepts, common question patterns, and practical strategies.
Reading Stanford CS329A
A lecture-by-lecture reading of Stanford CS329A on self-improving AI systems, grounded in materials attributable to each official session and paused where evidence is missing.
Cultivating Taste
Treating taste as judgment that can be observed, defended, and recalibrated, with a systematic practice for deciding what is worth making and what good work looks like when AI amplifies execution.
Product Builder Interview Prep
Preparing for product builder interviews across ten topics — product sense, metrics, strategy, execution, technical PM, growth, and AI product design. Each post focuses on one interview dimension with frameworks, case studies, and answer strategies.
AI Certification Prep
One preparation path per certification, built on the official exam guides: what each domain tests, which official material covers it, what to build, and the reasoning behind every schedule. Everything comes from official exam guides and certification pages — no exam-day accounts, no leaked questions.
Hermes Agent Documentation Guide
Reading Hermes Agent against the official Nous Research docs: install and upgrade, model providers and Nous Portal, the Tool Gateway, seven terminal backends, memory and skills, tools and plugins, the gateway and scheduling, the security model, and migrating from OpenClaw. Each post keeps the trade-offs and failure modes and leaves command details to the docs.
The Cloudflare Edge Stack
Every piece needed to build a full application on Cloudflare’s edge, read one at a time: the Workers execution model, where D1, KV and R2 each stop being the right answer, the framework layer of Hono and OpenNext, then Workers AI bindings and the domain and native-module problems that show up at deploy time.
CS146S: Ten Weeks of AI-Native Development
Reading Stanford CS146S "The Modern Software Developer" week by week — agent internals, context engineering, skills and customization, codebase readiness, code review, security, background agents, team-scale adoption, and the software factory. Each post is grounded in the course material and verifiable primary sources.
Reading Stanford CS230
A lecture-by-lecture reading of Stanford CS230, Autumn 2025 — what was taught, what has changed since, and where it agrees or disagrees with the practice written up elsewhere on this site.
The Agent Production Line
Reading agents as a production line: where the concept ends, how model and harness divide the work, context and memory, enterprise cases, security, the protocol layer, and the three shapes of RAG.
Taiwan's Drone Industry, Taken Apart
Taking the drone industry apart into verifiable layers — from the industry map and the supply-chain gap, through endurance physics and flight-controller and radio-link source code, to Taiwan’s regulatory authority, procurement records and counter-drone deadlock. Every post starts from primary material.
Learning How to Learn
Auditing the evidence behind learning science alongside how generative AI is actually used — which practices hold up, which merely circulate, and what pen and paper still do better.
Browser Automation and MCP
The routes for putting a browser in an agent’s hands: the trade-offs between the Playwright, Puppeteer and Chrome DevTools MCP servers, vision-driven Midscene, and how the CLI agents differ in what they can drive natively. Focused on where each route breaks.
AI Agent Systems in Practice
A practical series on AI agent systems, covering context, harness design, workflows, and multi-agent collaboration.
Claude Code Automation Guide
A practical series on Claude Code workflows, including hooks, skills, remote agents, routines, and team-scale automation.
Reading the OpenClaw Docs
Reading the 300+ official docs of OpenClaw, a self-hosted AI gateway, across 32 posts — installation and platforms, model providers, the agent runtime and memory, 24+ chat channels, sandboxing and threat model, tools and automation, gateway operations, plugins, and the user interfaces.
Building NobodyClimb
A climbing-community product written up end to end: positioning, why it needed AI at all, the system architecture, and the RAG pipeline. The technique-level potholes live in the RAG compendium; this series is about how the decisions got made.