IDVORKIN-AI-TOOLS

Explainers, field maps, and a few toys

Everything with a live page in this org, in one place. Field maps take a course, an author or a subject and lay it out page by page with the sources pinned; explainers take one question and answer it properly. Built by Igor Dvorkin's AI tools, reviewed by Igor. Cards say when each site was last touched. There are a few different kinds of explainer here; that page says what each kind is for and shows an example we think is good, and the production strategies page pulls out the craft moves so they can be mixed with any kind.

AI, models and safety 9 maps

Robert Moses Atlas — The Power Broker's New YorkForty-four years of appointed power on a map of New York: scrub from 1924 to 1968 and watch the parkways, bridges, beaches and expressways appear, along with the offices he stacked and the neighborhoods it cost. Six chapters, every card sourced.ATLAS2026-09 Entropy, Mapped — 3Blue1Brown on CompressionGrant Sanderson's Compression is Intelligence series in twenty pages: surprise, prefix codes, entropy, cross-entropy and KL divergence, ending at the loss function every language model is trained on. Wordle and Hamming codes included, every claim timestamped into the video.FIELD MAP2026-09 CS336, Mapped — Language Modeling from ScratchStanford's Spring 2025 course: seventeen lectures, each with a deep dive linked into the video by timestamp, a cleaned transcript, the slides and assignment, and verified supporting materials.FIELD MAP2026-09 Let's build GPT, MappedKarpathy's two-hour lecture as eight explainer parts linked into the video by timestamp, the code annotated line by line, a cleaned transcript, and his four exercises expanded.FIELD MAP2026-09 micrograd, MappedKarpathy's first Zero-to-Hero lecture, backprop from scratch, as eight explainer parts linked into the video by timestamp, the notebook code annotated, a cleaned transcript, and the exercise.FIELD MAP2026-09 Technical AI Safety, MappedBlueDot Impact's course: six units, twenty-seven chapters, every reading linked back, every exercise written out. Original explanations, not a copy of the course.FIELD MAP2026-09 AI Safety — A Field MapA technical explainer for engineers entering alignment, interpretability, evals and governance: what the subfields are, who works on what, where to start.FIELD MAP2026-07 HarmBench, MappedThe standard automated red-teaming benchmark, laid out so you can browse it by harm: a page for each of the seven categories, the four behavior types and their three grading paths, eighteen attack methods with a deep link per method, and what the ASR number hides. All 510 behavior IDs are permalinked and line-linked to the source CSV. Counts parsed from the shipped CSVs; four paper/repo discrepancies found and shown.FIELD MAP2026-09 Red Teaming Models — A Ramp-Up MapA self-contained guide for security professionals moving into adversarial AI testing: threat models, attack families, evals, and how the field is organised.FIELD MAP2026-06 When AirPods Eat Your WordsA real speech-recognition debugging case study: how a Bluetooth AirPods microphone destroyed on-device ASR, shown with the actual audio evidence.EXPLAINER2026-07

People, mapped 2 maps

The world, explained 5 pages

Tooling and engineering notes 2 pages

Trips, toys and samples 4 pages

Every site here is its own repo under github.com/idvorkin-ai-tools; the source, the build notes and the PR discussion are one click from any page. Missing something? Open an issue on this index's repo.