AI Foundations
Understand what AI is, how systems learn, where they fail, and how to judge claims without hype.
Free AI library
1019 plain-English guides, structured learning paths, and an open library — built by an independent 501(c)(3) nonprofit so anyone can understand modern AI.
Start here
Each course includes explicit outcomes, mapped competencies, practice activities, and an applied capstone.
Understand what AI is, how systems learn, where they fail, and how to judge claims without hype.
Use AI productively while protecting privacy, checking outputs, and preserving human accountability.
Evaluate workplace use cases, run safe pilots, measure value, and communicate changes responsibly.
Analyze AI systems through rights, equity, governance, safety, and public-interest outcomes.
Understand language models, retrieval, agents, evaluation, cost, and deployment safeguards through practical system design.
Topic tracks
Jump into the area you care about. Every track has multiple plain-English guides.
Full library
1019 of 1019 guides shown. Filter by track or search above.
AI colorization adds plausible, realistic color to black-and-white photos and film by predicting hues from grayscale patterns.
ApplicationsAI helps recover damaged, faded, or ancient documents by enhancing faint ink, reconstructing missing text, and even reading scrolls too fragile to open.
Audio AIMelGAN is a fully convolutional GAN-based vocoder that turns mel-spectrograms into raw audio waveforms in a single fast forward pass.
Audio AIUnivNet is a GAN vocoder that judges generated audio using multiple spectrograms computed at different STFT resolutions, sharpening high-frequency detail.
Audio AIDiffWave is a diffusion-based vocoder that synthesizes audio by iteratively denoising random noise into a waveform, conditioned on a mel-spectrogram.
FundamentalsLength normalization adjusts preference-tuning objectives so models stop winning approval just by writing longer answers.
TechnicalSpeculative RAG speeds up and sharpens retrieval-augmented generation by having a small, fast model draft multiple candidate answers from retrieved…
TechnicalBlock-sparse and native sparse attention let transformers attend to only the most relevant chunks of a long sequence instead of every token, slashing…
SocietyModel extraction attacks let an adversary clone a proprietary AI model just by querying its public API and training a copycat on the answers.
TechnicalWatermarking embeds a hidden statistical signal into AI-generated text so it can later be detected as machine-written, without changing what a human reader…
FundamentalsThe Bradley-Terry model is a century-old statistical method for turning pairwise comparisons (A beats B) into numeric scores.
SocietyReward hacking is when an AI maximizes its reward signal in unintended ways instead of doing what designers actually wanted.
Check what you learned with topic quizzes, then explore our structured courses and current certification requirements. Core guides remain free to read.