Learning Shelf — October 20, 2026

29 things to actually work through this month, split across courses, tutorials, references, papers, books, practice sites and lectures for AI, development and IT. Roughly half are the resources everyone already recommends; the other half showed up or changed in the last six months.

🎓 Courses & Curricula

CS50x: Introduction to Computer Science

Harvard's eleven-week survey starts at Week 0 with Scratch, then moves through C, Python, SQL, and the web stack before you've noticed the difficulty creeping up. It's been the default first course for self-taught programmers for over a decade, and the 2026 run adds a new unit on how AI tools change the actual work of writing software. Pricing: free to work through on OpenCourseWare or edX's audit track; score 70% or better on every problem set and the final project and you get a free Harvard certificate, or pay for a verified one on edX.

Practical Deep Learning for Coders

Jeremy Howard's course trains a working image classifier in lesson one and only explains the math once you have something running, which is the opposite order almost every other ML course teaches in. Part 1 is nine sessions of roughly 90 minutes each; a Part 2, covering diffusion models and deeper foundations in seventeen more lessons, picks up where it leaves off. Pricing: free, no certificate.

Machine Learning Crash Course

Google rebuilt this from scratch and the current version leans on interactive visualizations you can drag around rather than just watch, with new modules on large language models and AutoML folded into the roughly 15-hour self-study path. Start at linear regression; basic Python is the only real prerequisite. Pricing: free.

Introduction to Algorithms (6.006)

Erik Demaine, Jason Ku, and Justin Solomon's lecture recordings are from Spring 2020, and the material hasn't needed an update since: sorting, hashing, and graph algorithms work the same way they did then. Start at Lecture 1 and work the problem sets alongside the posted solutions. Pricing: free, open courseware, no certificate.

🧪 Interactive Tutorials & Playgrounds

Rustlings

A few hundred tiny Rust programs, most of which don't compile until you fix them, with rustlings watch rerunning the checker every time you save. It's what most Rust tutorials point you to after chapter two of the official book, and 64,000-plus GitHub stars later it's still the default. Pricing: free.

Learn Git Branching

A live commit graph sits next to a terminal, and typing real git commands (branch, rebase, cherry-pick) redraws the tree in front of you, which is the only way rebasing has ever actually clicked for me. Start with the Introduction Sequence; everything runs in the browser, no git install needed. Pricing: free.

freeCodeCamp

About 3,000 hours of curriculum across twelve certifications, from Responsive Web Design through a Full Stack Developer capstone, run by a non-profit with no premium tier anywhere in it. Start with Responsive Web Design if you've never written HTML. Pricing: free, including the certificates.

Transformer Explainer

Type a sentence in and an actual GPT-2 model runs in your browser, with sliders for temperature and top-k that change the next-token predictions as you move them. A Georgia Tech team built it, more than 125,000 people have used it, and it's the clearest way I've found to make attention weights feel like something physical instead of a matrix in a paper. Pricing: free.

📖 Reference & Cheat Sheets

MDN Web Docs

Mozilla has documented HTML, CSS, JavaScript, and the browser APIs since 2005, and it's the reference every working web developer has open in a tab without thinking about it. There's no real starting point here, since you tend to arrive on whatever API is giving you trouble, but the guided learning paths are worth reading start to finish if you're new. Pricing: free; MDN Plus is a paid add-on for search history and notifications, not for the documentation.

OWASP Cheat Sheet Series

More than 150 cheat sheets, each one tackling a single security topic (authentication, OAuth2, REST security, AI system risks), short enough to finish in one sitting and specific enough to apply the same day. Start wherever your current project needs it; there's no required order. Pricing: free, Creative Commons licensed.

Devhints

Rico Stacruz calls this "TL;DR for developer documentation," and that's exactly what it is: one Markdown page per tool, from Bash and vim to Terraform and Rails, with nothing padded in. Pick your tool from the index and start reading. Pricing: free.

Model Context Protocol Specification

The spec itself, not a blog post summarizing it, explains how hosts, clients, and servers talk over JSON-RPC to share context and tools between LLM apps. This revision landed 2026-07-28 and is the current one; start with the Architecture page before the protocol details get specific. Pricing: free, open access.

📄 Papers & Reading Groups

Attention Is All You Need

The eight-author 2017 paper that introduced the Transformer, and with it, nearly every language model built since. It's about eight pages of real content once you skip the abstract, and rereading it once a year is a decent check on how much of attention you actually still remember. Pricing: free on arXiv.

A Survey of Agentic Reasoning for Large Language Models

Twenty-nine authors mapped a research area that barely existed three years ago, organizing agents that plan, act, and improve from their own feedback into single-agent, self-evolving, and multi-agent sections. Posted in January 2026 and revised again in September, it's about as current a survey of agentic LLM work as exists right now. Pricing: free on arXiv, Creative Commons licensed.

Papers We Love

A community-run index of the computer science papers people bring to actual meetups, split across 80-plus folders covering distributed systems, cryptography, databases, and graphics, with local chapters in half a dozen cities. Pick the folder that matches what you're working on and start there. Pricing: free.

📚 Books & Long Reads

Crafting Interpreters

Bob Nystrom built two interpreters for the same toy language, one in Java and one in C, and wrote the whole thing up with hand-drawn diagrams so garbage collection and closures make sense because you implemented them yourself. Start at Part II if you already know how to code; Part I is a warm-up. Pricing: free online; the print and ebook editions are paid.

Deep Learning

Goodfellow, Bengio, and Courville's 2016 textbook is still what professors assign for the math underneath deep learning: linear algebra and probability first, then the architectures, then research-level chapters on generative models. Some of its examples have aged, but the foundations haven't. Pricing: free online; a print edition is paid.

Designing Data-Intensive Applications, 2nd Edition

Martin Kleppmann's second edition, co-written with Chris Riccomini and out since February 2026, updates the original's explanations of replication, partitioning, and consistency for the systems teams actually run today. If you've read the first edition, treat this less as a rewrite and more as the parts that needed catching up. Pricing: paid, print and ebook.

🧩 Practice & Challenges

Advent of Code

Daily puzzles released through December, solvable in any language, with every solution fast enough to finish on ten-year-old hardware. Start with the 2015 archive for the gentlest ramp, or jump into whichever year has an active leaderboard you want in on. Pricing: free; merchandise is sold separately.

Exercism

Eighty-five language tracks, each built from concept exercises and practice exercises, with volunteer mentors who review your code for free, which is the part that's genuinely hard to find anywhere else without paying. Pick a track and start with its first concept exercise. Pricing: free, forever, no premium tier.

Codewars

Community-submitted kata, ranked by difficulty and solved in-browser across 55-plus languages, with everyone else's passing solutions visible once you submit yours so you can see a cleaner or faster approach. Start with an 8kyu kata in whatever language you're learning. Pricing: free, with an optional Codewars Red subscription for extras.

CodeCrafters

Rebuild Redis, Git, Kafka, or SQLite from scratch in your own editor, pushing to git for feedback at each stage instead of working in a browser sandbox. It's the option for when reading about how Redis works stops being enough and you want to have built one. Pricing: free to try one challenge; full catalog access is a paid membership.

🎥 Talks & Lectures

MIT 6.S191: Introduction to Deep Learning

Nine lectures over two months each spring, each paired with a coding lab, open-sourced the same day they air in the MIT classroom. The 2026 run went from March 30 to May 25 and covered everything from sequence modeling to large language models. Pricing: free.

Neural Networks: Zero to Hero

Andrej Karpathy builds micrograd, then a character-level language model, then GPT itself, typing every line live and leaving the mistakes in when they happen. Eight lectures, about 20 hours total, and still the closest thing to watching someone who actually understands backpropagation explain it from scratch. Start with the micrograd lecture. Pricing: free.

Stanford CS25: Transformers United

A different researcher shows up most Thursdays each spring to talk through one piece of transformer research. This spring's sixth run covered everything from state-space model tradeoffs to serving transformers at scale in production. No Stanford affiliation needed; the Zoom livestream and recordings are open to anyone. Pricing: free.

🏅 Certifications & Paths

AWS Certified Solutions Architect – Associate

The AWS cert most hiring managers still mean by "AWS certified": architecture tradeoffs, cost optimization, resilience, and AWS's own Skill Builder courses and practice labs cover most of what the exam actually asks. Pricing: $150 to sit the exam; prep material is free.

CompTIA Security+

A vendor-neutral baseline that's stayed the standard entry cert for security roles for over a decade. CompTIA raised the exam fee to $439 in June 2026, and a V8 revision with more AI-risk content launches in November; Professor Messer's free video series still covers the current objectives end to end if you'd rather not pay for a course. Pricing: $439 for the exam; free study material exists elsewhere.

Certified Kubernetes Administrator (CKA)

A two-hour, hands-on exam where you fix broken clusters from the command line instead of picking multiple-choice answers, which is why it gets taken more seriously than most cloud certs. CNCF posts the curriculum breakdown and candidate handbook free on GitHub. Pricing: $445, includes one free retake.

AWS Certified AI Practitioner

AWS's newer foundational credential covers AI, ML, and generative AI concepts without requiring you to write code, sitting at roughly half the price of the Solutions Architect exam next to it. Pricing: $100; free prep through AWS Skill Builder.

---

29 things to work through this month. Every entry says whether it's free and whether it certifies. Found a dead link or have something to add? Reply to the email.