When you send a prompt to an LLM, every token figures out which other tokens matter to it. This lookup — called self-attention — is how transformers resolve references like “it”, connect verbs to subjects, and parse structure. Type any text below to see these patterns in action.