Tesla's FSD v14.3 Released
Tesla pushed Full Self‑Driving (Supervised) v14.3 with a rewritten AI compiler and revised reinforcement‑learning training that the company says cuts reaction times by about 20%. The update also targets tougher edge cases such as emergency vehicles and small animals, signalling Tesla’s continued cadence of incremental on‑road capability improvements. (x.com)
Tesla’s new driving update is about shaving fractions of a second off a car’s decisions. In Tesla’s own release notes for Full Self-Driving version 14.3, the company says a rewritten artificial-intelligence compiler and runtime cut reaction time by 20%. (notateslaapp.com) Full Self-Driving is Tesla’s driver-assistance system that steers, changes lanes, turns at intersections, and follows a navigation route, but Tesla says it is still supervised and does not make the car autonomous. Tesla’s support page says the driver must stay attentive and ready to take over at all times. (tesla.com) The basic job of the system is simple to describe and hard to do: cameras watch the road, software turns those images into a 3D picture, and the car chooses what to do next. Tesla’s owner manuals say the onboard computer uses camera inputs to process neural networks and guide the vehicle. (tesla.com) That is where reaction time matters. If a child runs toward a curb or a car suddenly cuts across a lane, the useful difference is not a new button or menu but how quickly the software sees the scene and commits to a maneuver. (notateslaapp.com) Tesla says version 14.3 also changed how the system learns from mistakes. The company’s release notes say it upgraded the reinforcement-learning stage of training, which is the part where the model gets rewarded for better choices, like training a dog with treats except the “treat” is a higher score in simulation and fleet-learned examples. (notateslaapp.com) The update is aimed at the ugly edge cases that make city driving hard. Tesla says version 14.3 improves responses to emergency vehicles, school buses, right-of-way violators, and small animals, and also improves traffic-light behavior at compound-light intersections and on curved roads. (notateslaapp.com) Tesla also says it upgraded the vision encoder, which is the part of the system that turns raw camera frames into something the driving model can understand. In plain English, that is the translator between pixels and road meaning, and Tesla says the new version is better in rare and low-visibility situations and better at reading traffic signs. (notateslaapp.com) Some of the changes are less dramatic but easier for owners to notice. Tesla says version 14.3 reduces unnecessary lane biasing and minor tailgating, makes parking-spot selection more decisive, and improves parking-pin prediction on the map with a “P” icon. (notateslaapp.com) This is rolling out on hardware version 4 vehicles in the United States, according to the release notes tracked from Tesla’s software update. Tesla also has a support page for a Full Self-Driving version 14 trial covering Model S, Model 3, Model X, Model Y, and Cybertruck owners in the United States, Puerto Rico, Mexico, and Canada. (notateslaapp.com) (tesla.com) The important part is that Tesla is still improving the system with many small releases instead of one giant leap. Version 14.3 is another example of that pattern: faster software underneath, more training on rare hazards, and another reminder from Tesla that the human in the seat is still the one responsible for the car. (notateslaapp.com) (tesla.com)
