AI Codex
Foundation Models & LLMsDevelopers

Attention Mechanism

Also: self-attention

The core innovation in modern LLMs — allows the model to weigh the relevance of every word against every other word in context, rather than reading sequentially.