Can a text-only model drive? Jev runs autonomous driving via a world model

jakedahn · x · 2026-09-22

Someone wired Jev — a model with no image inputs — to an unreleased world model used to train self-driving cars plus a local segmentation model, and let it drive autonomously. Result: 'Kinda? A little Crazy Taxi — but kinda amazing this works at all.' jakedahn finds the visual demos inspiring but notes Jev's lack of image input, saying things will get interesting once it goes multimodal.

Original post →

More from coding & agent

coding & agent channel →