← Back

Prompt to Response, what actually happens? And whats in the AI cake?

AI Summer School: Prompt to Response & Problems to Solve
Platform Calgary · July 2026

A dinner question answered in two seconds The trip to the data center and the API gateway What the question crossed on the way The question turned into tokens Prefill, when the model reads the whole prompt Decode, one token at a time Batching many questions on one chip The factory in layers, from user to power Where the AI build-out money goes Progress that looked like scale DeepSeek R1 and a different question What we can learn from DeepSeek Language is no longer the end goal From manipulating language to language as an interface Cheaper, controllable, useful reasoning Why this shift is happening now Prompt, model, answer replaced by a trajectory What we already do, and where each one breaks Six questions those approaches cannot answer yet Bottlenecks, and where open research can help