jevlike: an open reimplementation

The claim

Trains a small model to score a changing list of options in one forward pass, with Doom, chess, and Wikispeedia demos.

Our read

A good read for understanding how the architecture works.

Verified: The thing exists and the core claim checks out against a source you can open.

Test status

We haven't reproduced this yet. When we do, measured accuracy, latency, and cost will appear here next to the claimed figures.

Related entries