Unsupervised FSD in 2025 for Texas and California says Musk at Robotaxi / Cybercab reveal

[Darthamerica]

There has been 1.3 billion miles driven in FSD from Tesla alone, add in the other manufactures and you have plenty of data to determine if there are less crashes on average.

This article links to an independent study done by the University of Florida, that not only shows there is enough data, but that FSDis safer, with less accidents and fewer injuries during accidents due to reduced speeds pre accident(car acting faster than humans to brake and swerve)
https://www.linkedin.com/pulse/repo...ng-cars-safer-than-human-drivers-imagry-jfcrf

To many people, when they see 1.3 billion miles, they think wow a BILLION! Surely that’s enough. It’s not. Even Elon Musk acknowledges that publicly. First of all, those are Freeway/HWY miles. The most benign scenario requiring little more than lane keeping and TACC. Most of those miles aren’t useful for training. Tesla will need many billions of miles, orders of magnitude more, to have enough data to make such a comparison. And again, this isn’t even talking about city streets where you’re going to encounter the edge cases that the training needs to get to the point of equaling what humans can do.

This is a good start, but FSD is nowhere near the capability or safety of a human driver outside of narrow(AI) situations like road trips on relatively free flowing highways. Don’t underestimate what that brain in your skull is truly capable of. FSD has a long way to go!

Sponsored

 

[Woodrick]

You’re making my point. Rush hour and on highways are relatively easy for FSD. In fact that’s where most FSD mile data comes from. Human driven crash data comes from a wider range and more difficult conditions than FSD is capable of handling for now. It’s not an Apples to Apples comparison.

But if you said FSD is safer on the highway compared to humans alone on the highway, that’s probably closer to reality.

And Interstates, while they make up only about 2.5% of the roads, represent 25% of the roadway miles. I'm not sure how you discount them.
And not all Interstate miles are in stop and go traffic.

 

[Woodrick]

To many people, when they see 1.3 billion miles, they think wow a BILLION! Surely that’s enough. It’s not. Even Elon Musk acknowledges that publicly. First of all, those are Freeway/HWY miles. The most benign scenario requiring little more than lane keeping and TACC. Most of those miles aren’t useful for training. Tesla will need many billions of miles, orders of magnitude more, to have enough data to make such a comparison. And again, this isn’t even talking about city streets where you’re going to encounter the edge cases that the training needs to get to the point of equaling what humans can do.

This is a good start, but FSD is nowhere near the capability or safety of a human driver outside of narrow(AI) situations like road trips on relatively free flowing highways. Don’t underestimate what that brain in your skull is truly capable of. FSD has a long way to go!

Interesting, incorrect understanding of statistics.
For all of these political polls, do you think that each is polling 80,000 people? for each poll?

Because that's the same ratio of voters in the US (161 million) to poll respondents as US miles driven (2.6 trillion) vs your FSD miles (1.3 billion).

You may want to look at how statistics work, since this is truly a statistics issue.

The data is statistically valid.

 

[Darthamerica]

To many people, when they see 1.3 billion miles, they think wow a BILLION! Surely that’s enough. It’s not. Even Elon Musk acknowledges that publicly. First of all, those are Freeway/HWY miles. The most benign scenario requiring little more than lane keeping and TACC. Most of those miles aren’t useful for training. Tesla will need many billions of miles, orders of magnitude more, to have enough data to make such a comparison. And again, this isn’t even talking about city streets where you’re going to encounter the edge cases that the training needs to get to the point of equaling what humans can do.

This is a good start, but FSD is nowhere near the capability or safety of a human driver outside of narrow(AI) situations like road trips on relatively free flowing highways. Don’t underestimate what that brain in your skull is truly capable of. FSD has a long way to go!

Interesting, incorrect understanding of statistics.
For all of these political polls, do you think that each is polling 80,000 people? for each poll?

Because that's the same ratio of voters in the US (161 million) to poll respondents as US miles driven (2.6 trillion) vs your FSD miles (1.3 billion).

You may want to look at how statistics work, since this is truly a statistics issue.

The data is statistically valid.

This is not an incorrect understanding of statistics by me. It is by those of you who think this data validates your assumptions. If you understand statistics, you’ll understand that this FSD data isn’t accounting for all driving scenarios and drivers. It’s mostly highway driving whereas the non Tesla driving public data is inclusive of the full range of driving scenarios. That’s a huge difference. All miles aren’t created equal.

 

[Woodrick]

This is not an incorrect understanding of statistics by me. It is by those of you who think this data validates your assumptions. If you understand statistics, you’ll understand that this FSD data isn’t accounting for all driving scenarios and drivers. It’s mostly highway driving whereas the non Tesla driving public data is inclusive of the full range of driving scenarios. That’s a huge difference. All miles aren’t created equal.

Which report are you referring to that shows the safety data for 1.3 billion miles?
I believe that is based on Autopilot technologies, for which FSD is a part, but not whole.

The 2022 Tesla impact report (pg 77) did give a number for FSD engaged miles vs Autopilot. Autopilot was better than FSD, and both were better than the US average.

In made it to 1.3 billion miles in April 2024, but doesn't seem to have posted associated safety reports at that number.

Even just the basic car's safety features tend to be significant. Autopilot is over 10 times better than the average US driver.


2022 Tesla Impact Report.

Sponsored