News
Tesla replaces head of Autopilot with OpenAI’s neural net researcher
Tesla has replaced its Vice President of Autopilot Software Chris Lattner with Andrej Karpathy who will serve as the company’s new Director of AI and Autopilot Vision.
Lattner, a former 11-year Apple veteran who joined Tesla less than six months ago has departed the company, citing that the role was not a good fit him. “Turns out that Tesla isn’t a good fit for me after all. I’m interested to hear about interesting roles for a seasoned engineering leader!” says Lattner through his Twitter account. A Tesla spokesperson echoed the sentiment, telling Teslarati that “Chris just wasn’t the right fit for Tesla, and we’ve decided to make a change.”
Karpathy, who prior to joining Tesla worked at Google’s DeepMind and was a respected research scientist at Musk’s artificial intelligence OpenAI nonprofit, has extensive experience training neural networks and working with deep learning AI. “I like to train Deep Neural Nets on large datasets.” reads Karpathy’s Twitter profile.
In his new role as Tesla Director of AI and Autopilot Vision, Karpathy will report directly to Tesla chief Elon Musk, and work closely with chip expert and Vice President of Autopilot Hardware Jim Keller on advancing Tesla’s self-driving technology.
Tesla released the following statement regarding the hiring of Karpathy:
Andrej Karpathy, one of the world’s leading experts in computer vision and deep learning, is joining Tesla as Director of AI and Autopilot Vision, reporting directly to Elon Musk. Andrej has worked to give computers vision through his work on ImageNet, as well as imagination through the development of generative models, and the ability to navigate the internet with reinforcement learning. He was most recently a Research Scientist at OpenAI.
Andrej completed his computer vision PhD at Stanford University, where he demonstrated the ability to derive complex descriptions of images using a deep neural net. For example, identifying not simply that there is a cat in a given picture, but that it is an orange, spotted cat, riding on a skateboard with red wheels on brown hardwood flooring (http://cs.stanford.edu/people/karpathy/main.pdf). He also created and taught “Convolutional Neural Networks for Visual Recognition,” the first and still leading deep learning course at Stanford.
Andrej will work closely with Jim Keller, who now has overall responsibility for Autopilot hardware and software.
Elon Musk
SpaceX Starship Version 3 booster crumples in early testing
Photos of the incident’s aftermath suggest that Booster 18 will likely be retired.
SpaceX’s new Starship first-stage booster, Booster 18, suffered major damage early Friday during its first round of testing in Starbase, Texas, just one day after rolling out of the factory.
Based on videos of the incident, the lower section of the rocket booster appeared to crumple during a pressurization test. Photos of the incident’s aftermath suggest that Booster 18 will likely be retired.
Booster test failure
SpaceX began structural and propellant-system verification tests on Booster 18 Thursday night at the Massey’s Test Site, only a few miles from Starbase’s production facilities, as noted in an Ars Technica report. At 4:04 a.m. CT on Friday, a livestream from LabPadre Space captured the booster’s lower half experiencing a sudden destructive event around its liquid oxygen tank section. Post-incident images, shared on X by @StarshipGazer, showed notable deformation in the booster’s lower structure.
Neither SpaceX nor Elon Musk had commented as of Friday morning, but the vehicle’s condition suggests it is likely a complete loss. This is quite unfortunate, as Booster 18 is already part of the Starship V3 program, which includes design fixes and upgrades intended to improve reliability. While SpaceX maintains a rather rapid Starship production line in Starbase, Booster 18 was generally expected to validate the improvements implemented in the V3 program.
Tight deadlines
SpaceX needs Starship boosters and upper stages to begin demonstrating rapid reuse, tower catches, and early operational Starlink missions over the next two years. More critically, NASA’s Artemis program depends on an on-orbit refueling test in the second half of 2026, a requirement for the vehicle’s expected crewed lunar landing around 2028.
While SpaceX is known for diagnosing failures quickly and returning to testing at unmatched speed, losing the newest-generation booster at the very start of its campaign highlights the immense challenge involved in scaling Starship into a reliable, high-cadence launch system. SpaceX, however, is known for getting things done quickly, so it would not be a surprise if the company manages to figure out what happened to Booster 18 in the near future.
News
Tesla FSD (Supervised) is about to go on “widespread” release
In a comment last October, Elon Musk stated that FSD V14.2 is “for widespread use.”
Tesla has begun rolling out Full Self-Driving (Supervised) V14.2, and with this, the wide release of the system could very well begin.
The update introduces a new high-resolution vision encoder, expanded emergency-vehicle handling, smarter routing, new parking options, and more refined driving behavior, among other improvements.
FSD V14.2 improvements
FSD (Supervised) V14.2’s release notes highlight a fully upgraded neural-network vision encoder capable of reading higher-resolution features, giving the system improved awareness of emergency vehicles, road obstacles, and even human gestures. Tesla also expanded its emergency-vehicle protocols, adding controlled pull-overs and yielding behavior for police cars, fire trucks, and ambulances, among others.
A deeper integration of navigation and routing into the vision network now allows the system to respond to blocked roads or detours in real time. The update also enhances decision-making in several complex scenarios, including unprotected turns, lane changes, vehicle cut-ins, and interactions with school buses. All in all, these improvements should help FSD (Supervised) V14.2 perform in a very smooth and comfortable manner.
Elon Musk’s predicted wide release
The significance of V14.2 grows when paired with Elon Musk’s comments from October. While responding to FSD tester AI DRIVR, who praised V14.1.2 for fixing “95% of indecisive lane changes and braking” and who noted that it was time for FSD to go on wide release, Musk stated that “14.2 for widespread use.”
FSD V14 has so far received a substantial amount of positive reviews from Tesla owners, many of whom have stated that the system now drives better than some human drivers as it is confident, cautious, and considerate at the same time. With V14.2 now rolling out, it remains to be seen if the update also makes it to the company’s wide FSD fleet, which is still populated by a large number of HW3 vehicles.
News
Tesla FSD V14.2 starts rolling out to initial batch of vehicles
It would likely only be a matter of time before FSD V14.2 videos are posted and shared on social media.
Tesla has begun pushing Full Self-Driving (Supervised) v14.2 to its initial batch of vehicles. The update was initially observed by Tesla owners and veteran FSD users on social media platform X on Friday.
So far, reports of the update have been shared by Model Y owners in California whose vehicles are equipped with the company’s AI4 hardware, though it would not be surprising if more Tesla owners across the country receive the update as well.
Based on the release notes of the update, key improvements in FSD V14.2 include a revamped neural network for better detection of emergency vehicles, obstacles, and human gestures, as well as options to select arrival spots.
It would likely only be a matter of time before FSD V14.2 videos are posted and shared on social media.
Following are the release notes of FSD (Supervised) V14.2, as shared on X by longtime FSD tester Whole Mars Catalog.
Release Notes
2025.38.9.5
Currently Installed
FSD (Supervised) v14.2
Full Self-Driving (Supervised) v14.2 includes:
- Upgraded the neural network vision encoder, leveraging higher resolution features to further improve scenarios like handling emergency vehicles, obstacles on the road, and human gestures.
- Added Arrival Options for you to select where FSD should park: in a Parking Lot, on the Street, in a Driveway, in a Parking Garage, or at the Curbside.
- Added handling to pull over or yield for emergency vehicles (e.g. police cars, fire trucks, ambulances.
- Added navigation and routing into the vision-based neural network for real-time handling of blocked roads and detours.
- Added additional Speed Profile to further customize driving style preference.
- Improved handling for static and dynamic gates.
- Improved offsetting for road debris (e.g. tires, tree branches, boxes).
- Improve handling of several scenarios including: unprotected turns, lane changes, vehicle cut-ins, and school busses.
- Improved FSD’s ability to manage system faults and improve scenarios like handling emergency vehicles, obstacles on the road, and human gestures.
- Added Arrival Options for you to select where FSD should park: in a Parking Lot, on the Street, in a Driveway, in a Parking Garage, or at the Curbside.
- Added handling to pull over or yield for emergency vehicles (e.g. police cars, fire trucks, ambulances).
- Added navigation and routing into the vision-based neural network for real-time handling of blocked roads and detours.
- Added additional Speed Profile to further customize driving style preference.
- Improved handling for static and dynamic gates.
- Improved offsetting for road debris (e.g. tires, tree branches, boxes).
- Improve handling of several scenarios, including unprotected turns, lane changes, vehicle cut-ins, and school buses.
- Improved FSD’s ability to manage system faults and recover smoothly from degraded operation for enhanced reliability.
- Added alerting for residue build-up on interior windshield that may impact front camera visibility. If affected, visit Service for cleaning!
Upcoming Improvements:
- Overall smoothness and sentience
- Parking spot selection and parking quality
SpaceX Starship Version 3 booster crumples in early testing
Tesla FSD (Supervised) is about to go on “widespread” release
