Decoder. plain-English AI glossary

CLIP

● Core

OpenAIs model that learns to match image and text embeddings, enabling zero-shot classification and semantic image search without any training.

Think of it like

Learning that "dog photo" and "text: dog" are similar without ever being told "class 0 is dog."

Example

CLIP can classify images into 1000 categories without seeing a single example—just from the class names like "Labrador" or "Poodle."

How it actually works

Trained on 400M image-text pairs from web using contrastive loss: pull matching pairs close, push mismatched pairs apart. Image and text encoders output embeddings in shared space. Zero-shot via: embed new classes ("a photo of a {class}"), embed image, pick nearest class. Enabled by scale and diverse internet data.

For product teams

Massively reduces labeling burden; one model works for any classification task.

For engineers

Dual-encoder: image encoder + text encoder output to shared embedding space; contrastive loss during training; at test time, embed image and text, compute cosine similarity.

VLM — Foundation for.
Zero-shot — Enables.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome