Justin Schroeder: I rebuilt Tesla Full Self Driving with Jev in less than an hour. This model is a t

Tesla Full Self Driving (Jev demo) is now open source and you can try it online yourself! https://t.co/iS6Ch6EoI3 👈 try it https://t.co/CYLjbMDhu2 👈 keep it https://t.co/VCpsu8N61w

He didn’t fine-tune a driving model. He wired Jev into a Tesla-style FSD loop in under an hour. Justin Schroeder’s demo uses TypeSafe’s decision model — no text generation, just high-frequency choices — to steer a self-driving sim. They open-sourced it after. Try: https://t.co/ItJ6aIrLM3 https://t.co/zZjSjp0tiv

SOMEONE PUT JEV BEHIND THE WHEEL AND BUILT AN AUTOPILOT PROTOTYPE IN UNDER AN HOUR According to the developer, that was enough time to get a complete driving decision loop running in a simulation Turns, lane changes, intersections, and highway merges. For each maneuver, the model returns probabilities for possible actions and shows its confidence That is what sets it apart from conventional chat models. Feed it the current system state, and it returns a set of allowed actions with scores that code can use directly No lengthy explanations before every turn The experiment is still running in a virtual environment. But if this approach scales well, it could significantly lower the barrier to building autonomous driving systems