Part 1 of 8

Background

Inception Labs' about page lists nine papers under “Some of the technologies we’ve developed.” They aren’t a random reading list: they are the research trail of the company’s founders — Stefano Ermon, Volodymyr Kuleshov, and Aditya Grover — and together they tell one coherent story: how to make language models that generate text the way image generators paint images, instead of typing it out one word at a time. This page walks that story from the beginning. If you already know how LLMs generate text, skip to the second half of this section.

1. Deep background: what a language model does
2. Autoregressive generation: the typewriter
3. Meanwhile, in image generation: diffusion
4. Quiz: 5 questions