How AI agents work
An agent is an LLM in a loop — it reads a goal, picks a tool, acts, reads the result, and repeats until the job is done or a guardrail stops it.
How it works
Short pieces that open up one idea at a time. Read one when a lesson mentions something you want to see from the inside, rather than in passing.
15 explainers
-
-
How AI image generation works
Image models start from pure noise and remove it step by step, steered by your text prompt, until a picture that matches the words emerges.
-
How ChatGPT works
ChatGPT chops your message into tokens, scores every possible next token, and streams the winners back — shaped by training that taught it which answers people prefer.
-
How GPU inference works
Serving an LLM means one parallel pass over your prompt, then a one-token-at-a-time loop — kept fast by a cache, kept affordable by batching many users onto one GPU.
-
How Google Translate works
Modern translation reads the whole source sentence into numbers with an encoder, then a decoder writes the target language one token at a time, guided by attention.
-
How OCR works
OCR cleans up the photo, finds where the text lines are, reads each line as a sequence rather than letter by letter, and uses language knowledge to fix what the pixels left unclear.
-
How RAG works
RAG searches your documents first, then hands the best passages to the model so it answers from what it found instead of from memory.
-
How face recognition works
Face recognition finds a face, straightens it, compresses it into a list of numbers, and decides identity by measuring how close that list is to a stored one.
-
How fine-tuning works
Fine-tuning takes a model that already knows language, shows it your examples, and nudges its weights gently until it behaves your way without forgetting what it knew.
-
How model training works
Training is one loop run millions of times — guess, measure the error, trace blame backwards through the network, nudge every weight, repeat until the guesses get good.
-
How recommendation systems work
Recommenders log what you do, narrow millions of items to a few hundred candidates, score each one for you personally, and learn from what you click next.
-
How search ranking works
A search engine reads the web ahead of time into an index, understands your query, pulls thousands of candidate pages, and ranks them with hundreds of signals — all before you finish blinking.
-
How spam filters work
A spam filter turns each email into measurable features, scores it with a classifier trained on millions of labelled messages, and keeps learning as spammers adapt.
-
How speech recognition works
A microphone turns your voice into thousands of numbers per second, a spectrogram turns those into a picture of sound, and a model reads that picture into words.
-
How voice assistants work
A tiny always-on model listens for the wake word, speech recognition turns your request into text, an understanding layer picks the action, and synthesis speaks the reply.