FSD Beta 10.13

[jabloomf1230]

Am I missing something? Teslafi is only showing one install so far?

That's because new FSD beta firmwares are initially released to a subset of Tesla employees. There's apparently only one of those employees left using TeslaFi.

 

[aronth5]

That's because new FSD beta firmwares are initially released to a subset of Tesla employees. There's apparently only one of those employees left using TeslaFi.

Agree, plus Elon tweeted that "tweeking" was needed to 10.13 which adds a few days before the next group receives it. Seems like standard rollout so far

 

[Goose66]

Agree, plus Elon tweeted that "tweeking" was needed to 10.13 which adds a few days before the next group receives it. Seems like standard rollout so far

Right, so 10.13.1 or 10.13.2 for the rest of us. Probably in two weeks.

 

[Electroman]

Maybe there is still some ironing to do based the Cook's UPL testing that is happening in Cook's UPL - aka overfitting. This is called testing with production data.

 

[diplomat33]

Tesla Tidbits created a "safety score" for judging FSD Beta:

https://twitter.com/x/status/1538763209439596545

He basically substracts points based on the cause of each intervention. 100 is a perfect score. 0 is a terrible score.

A - Collision Avoidance (15 points)
B - Attempted an illegal act (10 points)
C - Confused or would have confused other drivers (3 points)
D - Attempted an incorrect act (2 points)
E - I was not confident with the car's action (1 point)
F - Perfect drives (2 point bonus)

Formula: FSDSS = 100 - (100(A+B+C+D+E-F)/Miles)

Looks like a pretty cool score system for judging FSD Beta.

 

[aronth5]

Tesla Tidbits created a "safety score" for judging FSD Beta:

https://twitter.com/x/status/1538763209439596545

He basically substracts points based on the cause of each intervention. 100 is a perfect score. 0 is a terrible score.

A - Collision Avoidance (15 points)
B - Attempted an illegal act (10 points)
C - Confused or would have confused other drivers (3 points)
D - Attempted an incorrect act (2 points)
E - I was not confident with the car's action (1 point)
F - Perfect drives (2 point bonus)

Formula: FSDSS = 100 - (100(A+B+C+D+E-F)/Miles)

Looks like a pretty cool score system for judging FSD Beta.

There are so many flaws with this scoring system it's hard to evaluate it's real value.

 

[Dewg]

Tesla Tidbits created a "safety score" for judging FSD Beta:

https://twitter.com/x/status/1538763209439596545

He basically substracts points based on the cause of each intervention. 100 is a perfect score. 0 is a terrible score.

A - Collision Avoidance (15 points)
B - Attempted an illegal act (10 points)
C - Confused or would have confused other drivers (3 points)
D - Attempted an incorrect act (2 points)
E - I was not confident with the car's action (1 point)
F - Perfect drives (2 point bonus)

Formula: FSDSS = 100 - (100(A+B+C+D+E-F)/Miles)

Looks like a pretty cool score system for judging FSD Beta.

It's penalized for avoiding a collision?

 

[diplomat33]

It's penalized for avoiding a collision?

No. It's penalized for a driver disengagement where FSD Beta would have collided if not for the driver disengagement.

https://twitter.com/x/status/1538763211423498241

 

[diplomat33]

There are so many flaws with this scoring system it's hard to evaluate it's real value.

How is it flawed? The more interventions, the lower the score, weighed by miles. Interventions are also weighed by severity which is good because interventions that are more safety critical should lower the score more than interventions that are less safety critical. The score also covers all the important types of interventions. And "perfect drives" improve the score.

Perhaps you don't understand how the scoring works?

Add number of interventions where FSD Beta would have hit something multiplied by 15
Add number of interventions where FSD Beta attempted an illegal act multiplied by 10
Add number of interventions where FSD Beta caused confusion in other drivers multiplied by 3
Add number of interventions where FSD Beta did an incorrect act by 2
Add number of interventions where driver was not confident in FSD Beta action but it was not illegal or unsafe multiplied by 1
Minus number of "zero intervention" drives greater than 3 miles multiplied by 2

That basically gives you a total points of "bad driving". Divide by total miles to get "bad driving" per miles. Then multiply by 100 and substract it from 100 to get a percentage of "good driving" per mile.

Here is an example:

https://twitter.com/x/status/1538763225210269696

 

[drtimhill]

Tesla Tidbits created a "safety score" for judging FSD Beta:

https://twitter.com/x/status/1538763209439596545

He basically substracts points based on the cause of each intervention. 100 is a perfect score. 0 is a terrible score.

A - Collision Avoidance (15 points)
B - Attempted an illegal act (10 points)
C - Confused or would have confused other drivers (3 points)
D - Attempted an incorrect act (2 points)
E - I was not confident with the car's action (1 point)
F - Perfect drives (2 point bonus)

Formula: FSDSS = 100 - (100(A+B+C+D+E-F)/Miles)

Looks like a pretty cool score system for judging FSD Beta.

It’s reasonable .. though how you judge some of those categories remains troublesome, especially C and D (I’m not even sure what would qualify as D). Also, there are lots of interventions caused by the car doing something dangerous that is not a collision and not illegal, which seems to either fall through the gaps or fall into a low point category (D/E)?

 

[diplomat33]

It’s reasonable .. though how you judge some of those categories remains troublesome, especially C and D (I’m not even sure what would qualify as D). Also, there are lots of interventions caused by the car doing something dangerous that is not a collision and not illegal, which seems to either fall through the gaps or fall into a low point category (D/E)?

Tesla Tidbits defines C and D:

https://twitter.com/x/status/1538763220080635905

 

[jabloomf1230]

Any standardized FSD beta scoring system would be helpful as long as it was based on thousands of drives by hundreds of drivers. IMO there is tremendous variability in any given driver perceiving these problems and subsequently disengaging FSD beta. My wife is more likely to disengage FSD beta in a specific circumstance than I am. And further, I sometimes prevented FSD beta from proceeding under questionable circumstances and as an afterthought I realized that FSD beta would have proceeded safely.

 

[diplomat33]

Any standardized FSD beta scoring system would be helpful as long as it was based on thousands of drives by hundreds of drivers. IMO there is tremendous variability in any given driver perceiving these problems and subsequently disengaging FSD beta. My wife is more likely to disengage FSD beta in a specific circumstance than I am. And further, I sometimes prevented FSD beta from proceeding under questionable circumstances and as an afterthought I realized that FSD beta would have proceeded safely.

Ideally, Tesla should collect data on all FSD beta drives and compute a total "safety score" for all the miles, from all the testers. Then we would have a score based on metrics measured the same way for everybody. Then we could see how the score changes from month to month or from version to version. That would be a more objective way to measure progress.

 

[Supcom]

Tesla Tidbits created a "safety score" for judging FSD Beta:

https://twitter.com/x/status/1538763209439596545

He basically substracts points based on the cause of each intervention. 100 is a perfect score. 0 is a terrible score.

A - Collision Avoidance (15 points)
B - Attempted an illegal act (10 points)
C - Confused or would have confused other drivers (3 points)
D - Attempted an incorrect act (2 points)
E - I was not confident with the car's action (1 point)
F - Perfect drives (2 point bonus)

Formula: FSDSS = 100 - (100(A+B+C+D+E-F)/Miles)

Looks like a pretty cool score system for judging FSD Beta.

All but B are subjective ratings. Therefore the "safety score" is little more than a measure of how closely FSD beta mimics the scorer's perception of how s/he drives.

 

[LowlyOilBurner]

If FSD b was tied to the “safety score” we all
love, it would record a score of like 6 based solely on hard braking and aggressive turning. Maybe throw unsafe following in the mix too.

 

[jeremymc7]

Rollout of beta to existing testers this weekend ?

 

[BBTX]

Rollout of beta to existing testers this weekend ?

My guess: It's remotely possible that a select few public testers might get 10.13 or 10.13.1 this weekend, but not certain. Most existing will probably get 10.13.1 or 10.13.2 over the next couple of weeks.

 

[jeremymc7]

Boo !!!

 

[JulienW]

Rollout of beta to existing testers this weekend ?

0 chance. First it will be released to a few YouTub/influencers and after a few days if all goes well it will start being SLOWLY rolled out over a couple of weeks checking for bugs. Also since Elon Tweeted that 10.13 "Still needs a few tweaks" that likely means there will be a 13.1 or at least some delayed before it goes to the YouTub/influencers. So probably still 2 to 4 weeks before most of us get it.

 

[diplomat33]

All but B are subjective ratings. Therefore the "safety score" is little more than a measure of how closely FSD beta mimics the scorer's perception of how s/he drives.

The only subjective metric is E. Collisions are not subjective. Unsafe, illegal and incorrect acts are not subjective.