Any chance latest beta is not implemented the same way in every car?

[cbdoc]

Reading through many threads and watching videos, I’m starting to think the latest beta is not performing/implemented the same way on every car. This seems totally implausible, but it’s just as hard to explain the dichotomy in feedback to this latest release.

Mine feels like a major step backwards on surface streets (worst performing since I’ve been in Beta for a couple of years), while highway it is incredibly good.

Examples of poor performance on my car 2022 MXP:

• Tons of phantom braking
• Right turns into random streets
• Merging randomly in and out of lanes and non-lanes
• Turn signals randomly going off very frequently

I’ve recalibrated and cleaned cameras but that hasn’t helped at all.

 

[Supcom]

Reading through many threads and watching videos, I’m starting to think the latest beta is not performing/implemented the same way on every car. This seems totally implausible, but it’s just as hard to explain the dichotomy in feedback to this latest release.

Mine feels like a major step backwards on surface streets (worst performing since I’ve been in Beta for a couple of years), while highway it is incredibly good.

All cars with the same software sets have the same programming. Differences in reported impressions are most likely due to variations in expectations, risk tolerance, and the car's ability to respond to each person's local driving environment.

Aside from obeying the traffic laws, much of our impressions of how well FSDb performs is highly subjective and based on how closely it matches our own driving.

 

[diplomat33]

Reading through many threads and watching videos, I’m starting to think the latest beta is not performing/implemented the same way on every car. This seems totally implausible, but it’s just as hard to explain the dichotomy in feedback to this latest release.

Mine feels like a major step backwards on surface streets (worst performing since I’ve been in Beta for a couple of years), while highway it is incredibly good.

No, the software is not different. The reason for the differences is relatively easy to explain. Driving is influenced by many factors (road condition, lighting, weather, traffic conditions etc...). And Tesla trains FSD Beta on certain data. If you are driving in an area or conditions that are different from the trained data, FSD beta may behave differently. For example, Tesla may use a lot of data from CA to train their NN. If you are driving in a small town in North Carolina where the roads are different than LA, the traffic conditions are different etc... then FSD Beta may drive differently since it was not trained on how to drive in NC. There are also different driver expectations, some want the car to drive more assertively, others want to more caution.

 

[IronCoffee_Max]

Also green found that FSDb definitely uses Tesla map data - and navigation - to perform better. If you're just driving with FSDb without navigation, it's going to do more poorly than with.

https://twitter.com/x/status/1641450306977562625

 

[cbdoc]

I understand

No, the software is not different. The reason for the differences is relatively easy to explain. Driving is influenced by many factors (road condition, lighting, weather, traffic conditions etc...). And Tesla trains FSD Beta on certain data. If you are driving in an area or conditions that are different from the trained data, FSD beta may behave differently. For example, Tesla may use a lot of data from CA to train their NN. If you are driving in a small town in North Carolina where the roads are different than LA, the traffic conditions are different etc... then FSD Beta may drive differently since it was not trained on how to drive in NC. There are also different driver expectations, some want the car to drive more assertively, others want to more caution.

I understand the software is the same, but the cars, hardware specifics, even underlying data (eg camera calibration) is different. I can upgrade to Windows 11 or Linux distro on 100 different machines and can expect differences on how they perform. But perhaps variability is much less on these cars. Not sure.

The regional differences I can buy to some extent, but very similar scenarios I would expect similar experiences.

 

[TresLA]

The wildly varying reports and posts is a common theme throughout the history of AP updates. Unless it was a software version that was pulled or updates to the fleet were stopped, there doesn’t seem to be a uniform experience.