Tesla FSD Review update for 10.9, with 3 crash interventions at one intersection

Topic: 
Tags: 

Tesla FSD's 10.9 update did offer some improvements, but along a new route I quickly ran into 3 different times I had to intervene to prevent a crash at one intersection, so it still remains "F" grade quality (and really far below.)

You can read the transcript and other text on Forbes.com at Tesla FSD Review 10.9

You can also check out The original review and its FAQ as well as my Video on why Tesla is crazy not to use HD maps.

FAQ

What made it try to jump into the intersection?

I was stopped at the light, but even though I had backed up, it could only see part of the crosswalk and not the stop line which was at the nose of my car. You can see on the map it's not aware of the line. I am told that since I reactivated FSD in that position, with no line, it wants to go, having received a sort of command to go, even though it sees the lights are red, and even though it also clearly sees the cars turning left into the lane to my side in its way. If I had pressed the pedal it might explain how it wanted to go, but just re-activating the system should not trigger this.

Comments

I always enjoy reading what others are encountering on FSD updates. I've been an early FSD beta customer since almost the beginning, and I have also witnessed the progression of the software. Unfortunately, this "new" article from Forbes is testing an outdated version of FSD. While I'm noticing improvements in the latest 10.10.2 FSD version, why -- in this digital age -- is this article published when it's already out of date?

I just got 10.10 a couple of nights ago, after the video was made. I have yet to drive with it. A large fraction of drivers don't have it yet. That's a pretty rough standard of "out of date!"

I will hand it to Tesla for being quick with releases, but unless people dedicate a lot of resources you won't get reviews published the moment the new releases come out.

The "starting to believe his own bull----" comment on Reddit is interesting.

Regarding your comments about Tesla and maps, the environment changes again as Bosch announces today it will acquire 3D mapping startup Atlatec.

Mobileye is probably not to happy that Bosch and Atlatec are imitating Mobileye mapping solution to sell to VW.

They may or may not be happy, but competition is probably good for the world. MobilEye has a lot of other assets, including being Intel and making processor and Lidar and Radar chips. Being in one of the top processor companies is a good position.

When you succumb to geofencing, you're no longer AI. That's not what Musk is after.

It's plausible to think that you may need to build a general AI and apply it to a task rather than use all and any means to solve it directly.
However when you look at the rate that FSD is improving, or not, it doesn't much look like Tesla will achieve it any time soon.

Come now, there's lots of problems solved without general AI, in fact every problem ever solved to date!

It is often joked that "Once we figure out how to do it, that's not longer AI." Humans need extra information about the road to drive, at least to drive fast. That's why there are lines on the road and signs telling you a curve or merge is coming. That's all the map is, more information about the road that the computer needs, which is different from what the human needs. They both need it.

akpthy taking 4 month sabbatical according to elon tweet

Andrej has tweeted that he's really just taking a break. History at Tesla says many don't return. Can understand why he needs one. But this is also not a time that Tesla will want to be without it's AI head, so who knows what it means?

the 4 months of rest leaves an impression no one is in a rush to declare fsd is rounding third base.

Add new comment