The Information Machine

"You could basically imagine, completely neural computers in a certain sense. Imagine a device that takes raw videos or …

Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-05-23

Andrej Karpathy, quoted by Rohan Paul, envisions a future computing paradigm in which neural networks process raw video and audio directly and diffusion models render unique, context-specific UIs in real time.

Open original ↗

Appears in

Extraction

Topics: neural-computingdiffusion-modelsfuture-ai-systemsui-generation

Claims

  • Future computing devices could process raw video or audio input directly through neural networks, bypassing traditional software stacks.
  • Diffusion models could be used to render UIs that are dynamically unique to each moment and context.
  • Karpathy envisions this as a transition toward 'completely neural computers.'

Key quotes

You could basically imagine, completely neural computers in a certain sense. Imagine a device that takes raw videos or audio into basically what is a neural net, and uses diffusion to render a UI that is unique for that moment in a certain sense.