Decoder. plain-English AI glossary

VLM

● Core

Neural network that understands both images and text—can caption photos, answer questions about images, or generate images from descriptions.

Think of it like

A person who can read, write, and understand pictures—richer than someone who only reads or only looks.

Example

GPT-4V takes an image and asks "what is in this photo?" and responds "a beach sunset with people swimming"; CLIP retrieves images matching text queries.

How it actually works

Vision encoder (ResNet, ViT) extracts image embeddings; text encoder converts words to embeddings; both sit in same space for alignment. Training requires image-text pairs (billions from web-scraped captions). More powerful than separate vision and text models but more compute-intensive.

For product teams

Enables richer user experiences: visual search, image Q&A, content understanding for moderation.

For engineers

Two-tower architecture: image tower outputs embeddings, text tower outputs embeddings; contrastive loss aligns them; downstream tasks use aligned space.

Related

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 DecoderAdd to Chrome