In plain English
Early language systems counted words. Embeddings later turned words and images into lists of numbers where similar things sit close. Vision tasks include classification, detection, segmentation and reading text from pictures. Modern models handle text and images together, but simple tools are often enough.
- Embedding
- Numbers that stand for meaning, placed so similar things are near.
- Object detection
- Finding objects and drawing boxes around them.
- OCR
- Reading text from a picture.
Picture it
Two ways to read the world
Modern multimodal models handle both in one system.
Language
- Text becomes tokens
- Embeddings capture meaning
- Tasks: classify, find names, translate
Vision