Resources+
AI Learning GuideAI Circle Human Data LandscapeMarket maps+
All market mapsHuman Data for Physical AIPhysical AI AtlasMultilinguality MapCoding AtlasTrust & Safety MapExpert Data MapRL Environments AtlasEvals & Benchmarking AtlasAudio, Image & Video DataData Licensing & Private DataLabeling Platforms & Data WorkforcesField notes from the operational frontier.
Deep dives on evaluations, post-training, human data, infrastructure, and the systems that decide whether AI actually works.
Start here
New to AI, or trying to keep up? The AI Learning Guide is what our researchers and operators send when someone asks where to begin. Pick the path that fits you, or read it end to end.
I’m new to AI
Start with the handful of papers and books that explain modern AI.
Start with the basics → 02I want to keep up
One podcast, one newsletter, and the people worth following.
Build the habit → 03 with
I want a certificate
Courses and credentials employers recognize. Free Data Gradient certificates are included.
See courses → 04I want to know who’s building
A filterable map of 140+ companies across the AI stack.
Explore the landscape →From the learning guide
Three picks from each part of the guide. Every list goes deeper on the full page.
Free certificates · Data Gradient
Get certified in AI data work
Free courses and certificates for AI training and evaluation work.
Browse courses ↗
Maps
Interactive market maps of the companies behind AI’s human data. Each one is searchable, filterable, and open to suggestions.
01 · Market mapHuman Data for Physical AI
Companies providing human data for physical AI. Grouped by data type and limited to those with public funding above $10M.
Explore the map →
02 · Interactive atlasPhysical AI Atlas
Search, filter, and shortlist human data vendors for physical AI across egocentric, teleoperation, synthetic, and more.
Open the atlas →
03 · Interactive mapMultilinguality Map
Companies selling multilingual human data for AI, across localization, evals, speech, RLHF, and low-resource languages.
Open the map →
04 · Interactive atlasCoding Atlas
Companies selling coding data for AI, across code SFT, RLHF and code evaluation, benchmarks, agentic environments, and security.
Open the atlas →
05 · Interactive mapTrust & Safety Map
Companies selling trust and safety data and services for AI, across moderation, red teaming, safety evals, alignment data, and guardrails.
Open the map →
06 · Interactive map · NewExpert Data Map
Companies selling expert human data for AI, across STEM and reasoning, medical, legal, finance, and expert talent networks.
Open the map →
07 · Interactive atlas · NewRL Environments Atlas
Companies building RL environments and agent data, across gyms, enterprise workflows, computer use, tool use, verifiers, and sandboxes.
Open the atlas →
08 · Interactive atlas · NewEvals & Benchmarking Atlas
Companies evaluating AI, across eval platforms, observability, benchmarks and leaderboards, audit firms, and certification bodies.
Open the atlas →
09 · Interactive map · NewAudio, Image & Video Data
Companies selling multimodal data for AI, across speech, voice talent, audio labeling, licensed libraries, captioning, and 3D.
Open the map →
10 · Interactive map · NewData Licensing & Private Data
Who buys and sells private data for AI, across data buyers, licensing brokers, company records, licensed text, and consumer data.
Open the map →
11 · Interactive map · NewLabeling Platforms & Data Workforces
Companies selling labeling, by how you buy: annotation software, open-source tools, managed vendors, BPOs, and crowd marketplaces.
Open the map →Blog
Field notes and deep dives on evaluations, post-training, human data, and the systems that decide whether AI actually works.
Latest · Kimi K3Kimi K3 Just Changed The Way AI Models Are Built
Chinese AI labs just caught up in a big way. Kimi K3’s architecture achieves 2.5x training efficiency with impressive memory recall.
Read the brief →
Deep dive No. 02 · Meta FAIRYour brain just became an LLM.
Meta’s TRIBEv2 predicts how video, audio, and text drive human brain activity. It then uses that to reproduce classic neuroscience experiments.
Read the deep dive →
Edition No. 01 · Multilingual evaluationYour AI agent isn’t wrong. Your benchmark is.
What a dinner with LILT’s CEO revealed about a benchmark error distorting how we evaluate AI in every language but English.
Read the brief →Notes from the rooms, in your inbox.
What researchers are working on in post-training, evaluation, and human data, plus the next rooms near you.
Free. Unsubscribe anytime.
Get the guide as a PDF
Enter your email to unlock the download. We will only use it to let you know when the guide is updated.