Show HN: universal application where LLM does all computation directly (opens in new tab)

Universal is a demo of a universal application ("OS") where all computation is done by the LLM. This is very similar to the idea sketched by Karpathy in his recent YC AI SUS talk.

This is based on a loop where user commands, or mouse clicks, are fed to the LLM, and the LLM is instructed to simply render the next frame, as if it was rendering the frame of a video. In this particular case, we actually render static HTML+CSS as the "image" because image output from existing LLMs doesn't have high enough text fidelity.

Computation is done by the LLM itself, the LLM does not "write code" it IS the code.