Eventually accidents are going to happen and the author's projection from human statistics and a simple relative safety multiplier is not a bad rough model, but there are reasons to believe that the distribution of FSD accidents could be much more heavily weighted towards minor events than human accidents are. The failure modes for FSD are going to be different from the way that humans make mistakes and that's going to change the character of accidents. Human accidents are dominated by attention failures. Whether due to illness, impairment, distraction, or just the human limitation of only being able to see in one direction at a time, human accidents involve a lot of events where the driver(s) didn’t see the accident coming, or didn’t see it until very late. In those accidents the impact velocity is sometimes very high. FSD will have fewer of these failures and as a result will be able to take action sooner and reduce impact velocities further than the average human accident. FSD also has faster reflexes and will employ better brake modulation than the average human. Additionally FSD is less likely to engage in high speed driving than more adventurous human drivers. All of these things will tend to reduce the average impact velocity when FSD is driving. Because accident severity follows impact energy, and impact energy goes up with the square of impact velocity, even moderate reductions in impact velocity can have a big impact on the severity of an accident and we should see that in the accident statistics. There’s an irreducible floor on accident rates. Even perfect driving on the part of one vehicle isn’t perfect protection on roads that other drivers share. This floor is probably something like 10x reduction relative to the human rate of accidents and supervised FSD, currently at about 7x reduction, might already be getting into that territory. But the nature of the accident matters to severity and the different characteristics of FSD as a driver give hope that the rate of minor incidents won’t be linearly predictive of the rate of serious incidents. We might know soon if this is true. Once we have data on 5M unsupervised miles we can set a 95% confidence rate for minor incidents. At 50M we’ll be able to say something about serious incidents. There are going to be accidents. And it’s possible that FSD ends up being “only” 10x better than human drivers. This will save a lot of lives and avoid enormous misery and waste. But in a world where the knives are coming out for anyone who threatens incumbent business models *any* accident will be an opportunity for opponents to smear FSD. Still, the jury isn’t out on the possibility that FSD ends up being so dramatically safer than human drivers that opponents have to seriously demean themselves.

Any chance better parking logic (especially at Superchargers) is planned for the next FSD v14 update @aelluswamy? I hoped sitemaps would help in situations like the one below however it currently doesn't seem to make a difference. 🙏

“You shouldn’t drive like that. That’s so unsafe. If your airbags go off then your leg will be shoved behind your head.” That’s literally every Friday night for me. I am not afraid of that.

Last week I purchased a motorcycle dolly for my Hypermotard and the box had 2 big holes in it. After opening the box, it was quickly apparent I was missing a majority of the bolts, washers, nuts, etc. Amazon support here I come...

the current fsd streak is good at measuring "of the intentionally fully autonomous drives, all of them ended successfully enough" it is not good at measuring "100% of vehicle control was autonomous" that's okay at this stage, but we need to quickly get to the latter

Such a bad take. Clearly @Tesla Model S & X were a success that lead to everything that has followed. Those cars highlighted that EVs would work, were cool & a game changer for the industry. The Model S shocked the auto industry & auto journalists. There is no Supercharger Network, M3, MY, Cybertruck or FSD without those cars. ✌️

We Beat the Autonomous Cannonball Record by 8 Hours Using Tesla FSD v14.3.2! Introduction On the morning of May 8th, myself, @BLKMDL3, and @AaronS5_ pulled out of the legendary Red Ball Garage in Manhattan with one goal in mind: beat the fully autonomous cannonball record of 58 hours and 22 minutes set by Alex Roy and his team back in January. This was our first cannonball attempt, so we knew we would face plenty of challenges along the way, but we figured it would be a blast to try at least once. Worst case, we would get to see parts of the country a few of us had not yet seen and come home with some great stories. Best case, we would etch our names into cannonball history. Spoiler alert: it went better than any of us could have imagined! Criteria Before diving into the trip itself, it is important to define exactly what counts as an intervention and what behavior was explicitly against the rules. The zero-intervention claim only means something if the criteria behind it are clear. Alex Roy did a fantastic job defining much of this criteria during his record-setting runs, drawing on his time at Argo AI, where safety operators were trained on what a real intervention actually is. We felt his framework was fair, honest, and consistent with how SAE and the major AV companies classify disengagements, so we saw no reason to reinvent the wheel. Here is what we held ourselves to: - No accelerator, brake, or steering inputs of any kind. The car must control all three throughout the entire trip, and that includes parking at every charging stop. - No manually changing gears, including putting the vehicle into park. If the car needed to be in park, the car had to put itself there. - No unbuckling the seatbelt or any other tricks to preserve an FSD engagement streak. - No CAN bus devices or nag cheat devices of any kind. All driver monitoring systems remained fully active for the entire run. - Speed profiles could be switched by the human observer. FSD has no concept of it being in a "race," so adjusting the profile is equivalent to a passenger asking a chauffeur to drive faster or slower. The car still makes every driving decision. Simply put, if the car could not do all of the driving, we did not deserve the record. Vehicle Configuration - 2025 Model 3 with AI4 - FSD v14.3.2 - No CAN bus devices of any kind - No nag cheat devices or props to trick the driver monitoring system The car was completely stock from a driver monitoring perspective. There were no CAN bus devices plugged in anywhere on the vehicle, and no nag cheat devices or props of any kind to fool the cabin camera or the steering wheel torque sensor. Every layer of Tesla's driver monitoring remained fully active from coast to coast. Challenges Anyone who tells you a cross-country run is uneventful has never done one. Mother Nature threw plenty at us, starting with a flock of geese that decided the highway was a great place to cross. FSD identified and reacted to them without any drama, which is more than I can say for my own heart rate. Wildlife was a recurring theme throughout the run, with active deer and jackrabbits near the roadway at multiple points, especially during the overnight stretches. The human drivers around us proved to be a bigger wildcard than the animals. At one merge, a trucker made a genuine effort to run us off the road, and FSD handled it calmly while the three of us held our breath. Inattentive drivers drifting around us were a constant reminder of why this technology matters in the first place. The car never got tired, never got distracted, and never got angry, which is more than I can say for some of the humans we shared the road with. Weather and lighting added another layer of difficulty. We hit heavy rain in Missouri with minimal hydroplaning, and the car tracked confidently through conditions that would have had me white-knuckling the wheel in years past. Glare and direct sunlight created poor lighting conditions at numerous points throughout the trip, and of course we tangled with our fair share of traffic along the way. Through all of it, FSD v14.3.2 just kept driving. Results - 0 disengagements - 0 interventions - Official time: 49 hours and 55 minutes That beats the previous fully autonomous record by more than 8 hours, and the car did every single bit of the driving, including parking itself at every Supercharger stop along the way. For validation purposes, we recorded the entire drive and what you are seeing here is the sped up version of this footage. We will also be uploading 4 parts of the 1x footage which equates to over 41 hours of driving footage. We want this record to hold up to scrutiny, and the footage speaks for itself. Closing Thoughts This was truly a once in a lifetime experience, though who is to say we will not attempt it again down the road. Watching a consumer vehicle you can buy today drive itself across the entire country without a single intervention is something I will never forget, and sharing that moment with two great friends made it even better. We are excited to see more folks join the fun, and we wish @DavidMoss, @scotsrule08, and @OwenSparks the best of luck during their upcoming run. We will be following along on their journey and hope they are successful in going even faster. Records are meant to be broken, and honestly, nothing would make us happier than watching this one fall as the software keeps improving. Stay tuned!

After spending more time with FSD v14.3.4 over the last few weeks, I have noticed even more movement on the steering wheel when the car is unsure of a situation. This has been more prevalent than in other v14.3 releases. It's almost as if the car is thinking aloud. It's not a critical issue but I really hope they can zap this in a future build. I think we'd all prefer the car not to jerk the wheel back and forth when it's trying to reason and figure out how to proceed. Has anybody else experienced this behavior with v14.3.4, and if so, what situations triggered it?