Safety Score

[ConSaab]

It’s just strange, I’ve logged 270 miles yesterday with perfect 100. Today 95 miles with 100. According to spreadsheet, I should be rounding to 100 but it won’t budge.

 

[mwasylyk]

It’s just strange, I’ve logged 270 miles yesterday with perfect 100. Today 95 miles with 100. According to spreadsheet, I should be rounding to 100 but it won’t budge.

What does Safety Score Calculator say when you enter in all 15 days with the total days mileage and the score for the day? If you click on the info button in the top right and hover over the calculated score it will show the exact decimal place it is estimating you currently have

 

[ConSaab]

What does Safety Score Calculator say when you enter in all 15 days with the total days mileage and the score for the day? If you click on the info button in the top right and hover over the calculated score it will show the exact decimal place it is estimating you currently have

I used the spreadsheet… below is a screen shot..

 

[ConSaab]

I noticed that I had high percent (8.0%) unsafe following. If I plug my scenario in the simulator and reduce my unsafe following percentage to 2.0. It shows then I’ll get a 100 overall.

Maybe I shouldn’t have done all my 272. Miles yesterday on mostly AP. I should’ve manual driven. The freeways were pretty empty in my area.

 

[n.one.one]

The TeslaFi 100 scores have jumped again to 12% as of now and that's a jump from 11% about 29 hours ago.
Just for reference here are all of the screen captures I've posted since Thursday morning (2 days ago) with the most recent at the top. Note that their FSD fleet vehicle count that they report on has gone from 1,569 2 days ago (far below) to 1,880 just now.

 

[mwasylyk]

The TeslaFi 100 scores have jumped again to 12% now and that's a jump from 11% about 29 hours ago.
Just for reference here are all of the screen captures I've posted since Thursday morning (2 days ago) with the most recent at the top. Note that their FSD fleet vehicle count that they report on has gone from 1,569 far below to 1,880 just now.

When you say jumped are you trying to say there was a big batch update to the stats all at one time? It is possible it's just the way the TeslaFi reporting is setup to update their leaderboard. Or are you trying to say you think the Tesla API just start reporting a lot of new 100 scores for people at once? Or are you just sharing the newest stats lol.

 

[n.one.one]

When you say jumped are you trying to say there was a big batch update to the stats all at one time? It is possible it's just the way the TeslaFi reporting is setup to update their leaderboard. Or are you trying to say you think the Tesla API just start reporting a lot of new 100 scores for people at once? Or are you just sharing the newest stats lol.

I'm just reporting the newest stats as shown by TeslaFi.
I'm going to assume that this is almost real time because when I finish a drive my Safety Score and all other stats are updated immediately. The other stats include current battery level, miles driven on the last trip, odometer, etc. If my info is updated that fast there is no reason their overall database wouldn't update in the same way, including Safety Score stats.

 

[Iain]

How many cars are represented in Teslafi....I have no idea what it is.

 

[n.one.one]

How many cars are represented in Teslafi....I have no idea what it is.

TeslaFi currently tracks 16,365 subscribing vehicles. It's a stat nerds dream. See my signature for some information.

 

[drtimhill]

Pain and simple, I really think it was just easier for them to just have a score that gave them a metric(super accurate or not) about how safe of a driving environment the car is in to pick the early stages of the rollout. I think the phrasing that it's a safety score and somehow supposed to be representative of your driving skills is what is irking some people. Because a lot of the score at the moment is really just being affected by where you are driving and how able(situational) you are to drive ULTRA slow and careful. Locations where it's easy enough to get 100 are probably locations as a generalization they would be more comfortable handing out the beta, just to reduce risk at first.

I tend to agree, but behind it I smell the whiff of regulators. I think they are buying a different kind of insurance: they want to be able to defend how they chose the beta test drivers, and who can argue with basing it on an insurance-like factoring system based on how insurance companies worked for years?

Tesla also have to be VERY careful about choosing test drivers .. a lot of people here have said "choose one or two safe locations" or "choose very safe drivers". The trouble is, you are then building a car that can only drive in those locations/conditions, since the NN is trained with what is essentially a biased sample set. And in general AI doesnt work by just adding in other sample sets to "extend it" .. you really have to build and train from a comprehensive set early on (there are exceptions). If you want to recognize faces, you train with more and more faces, and the more you add, the more you get better results .. you dont train the NN to be really good at noses, and then add (say) eyes to the training sets. That doesnt work.

 

[drtimhill]

Elon if you are listening let’s throw another weight into the FSD beta have / have not equation… If you bought before March 2019 you automatically go to the front of the line (as long as you already have a rounded to 100 score ).

I agree but doubt it will happen. In general Elon/Tesla have been tone deaf to those who invested in FSD early.

 

[powertoold]

In general Elon/Tesla have been tone deaf to those who invested in FSD early.

It was also cheaper back then ($3k was nothing in the scheme of available ADAS features back then)

Also, back then, people got a $7,500 tax credit. I didn't get any federal tax credit. I should be complaining more!

I demand Elon give me FSD beta first because I got federal $0 tax credit!!!!

All these whiners don't understand how ridiculous they are.

 

[drtimhill]

Do explain your thinking here!

Are you saying people with 100 scores are the only ones doing two-button resets? I would guess it is more common in the 100 score group. But they are certainly not the only ones. So how do you reconcile that? How do you know that the 100 drivers present a greater risk? This is what you are saying.

I think the two-button reset trick is awful but I don’t think lopping off 100 scores is the ticket to fixing it. I’d suggest Tesla forcing a software update closing the loophole and ignoring future data which is not from the updated software.

OK, so how to you get 100% score without cheating?

-- By driving very little (lowers chance of mistake, and formula doesnt seem to factor in mileage so far as we know)
-- By living in a very quiet area with few/no traffic signals, hills, other drivers etc.

But is this the profile Tesla want/need? FSD isnt going to get tested on a car that sits in the garage 6 days out of 7. Nor is it going to get the challenges it needs in super-quiet areas with no other traffic or messy/busy intersections etc. Yet that is precisely the profile that the 100% score is selecting for. Or, you game the system, and do Tesla want those people as testers???

I agree just throwing away all 100% scores is a blunt way to do, and I was only ½ serious, but I do think they have picked an odd way to choose testers (though see my comment elsewhere about them doing it so they can defend the methodology to the DMV).

TBH I think the first wave of testers actually are selected for being the "granny" set, so they can let the car out with the absolute least chance of an issue. Sure, they get no data on interesting driving situations, but they can go to the DMV after a few weeks and say "look ... we have XXXX miles and no issues .. all those fears you had were unfounded."

Then they can move down to ppl who drive in normal (messy) conditions, where they can start the massive dataset gathering they need for the next phases of development.

 

[EVNow]

The TeslaFi 100 scores have jumped again to 12% as of now and that's a jump from 11% about 29 hours ago.
Just for reference here are all of the screen captures I've posted since Thursday morning (2 days ago) with the most recent at the top. Note that their FSD fleet vehicle count that they report on has gone from 1,569 2 days ago (far below) to 1,880 just now.

View attachment 719662

So, of the 311 new people who opted in - 84 of them have a score of 100.

Ofcourse that is assuming all the increase in 100 (225 from 141) comes from new opt-ins. Some of those are likely moving up from 99 and other lower scores.

 

[n.one.one]

.................. Some of those are likely moving up from 99 and other lower scores.

That would be my assumption.

 

[dburkland]

Yeah maybe they are combing through the data as we speak. That should limit the pool. Definitely a lot of people will be upset though.

As has been mentioned, there are occasional legitimate reasons for a reset so it’s tricky - a no tolerance policy is tough.

It’s also not clear exactly what data they could look at if the safety data is lost. I believe they can easily figure out how many unlogged miles exist and probably how many resets were done, but that is it.

Maybe they should spend all those cycles on you know, polishing the actual code in the release vs deciding who does / does not get it in this first limited release. Or simply publish strict rules for initial testers that they have the right will revoke

It was also cheaper back then ($3k was nothing in the scheme of available ADAS features back then)

Also, back then, people got a $7,500 tax credit. I didn't get any federal tax credit. I should be complaining more!

I demand Elon give me FSD beta first because I got federal $0 tax credit!!!!

All these whiners don't understand how ridiculous they are.

Keep in mind even with tax credit I paid much more for my 3 than you would today for a similar configuration. Also I wouldn’t consider it whining when he promised early access to early adopters numerous times. Either way FSD beta will arrive when it is ready and I won’t get my hopes up until then

 

[drtimhill]

It was also cheaper back then ($3k was nothing in the scheme of available ADAS features back then)

Also, back then, people got a $7,500 tax credit. I didn't get any federal tax credit. I should be complaining more!

I demand Elon give me FSD beta first because I got federal $0 tax credit!!!!

All these whiners don't understand how ridiculous they are.

no, lets me fair here. FSD is not transferrable, so there is a clock ticking as well. If you buy a car and keep it for 5 years, then the cost can be divided up over those 5 years if you have the feature. Sure FSD was cheaper then, but amortized over its useful lifetime it works out more for early adopters. I dont think that's whining.

 

[powertoold]

no, lets me fair here. FSD is not transferrable, so there is a clock ticking as well. If you buy a car and keep it for 5 years, then the cost can be divided up over those 5 years if you have the feature. Sure FSD was cheaper then, but amortized over its useful lifetime it works out more for early adopters. I dont think that's whining.

That's true, but comparing a $7,500 tax credit vs $0, FSD was free at $5k, $4k, $3k, and $2k... all the prices prior to 2019

 

[AWDtsla]

Searching through this thread, I couldn't come to a clear answer:
Does the car filter out gravity from it's accelerometer reading for braking and turning? Because it sure feels like it doesn't. Turning slowly around a high grade corner seems to ding aggressive turning.

 

[powertoold]

Keep in mind even with tax credit I paid much more for my 3 than you would today for a similar configuration

Yup but you also got your 3 earlier than others, and got more tax credit, and got to enjoy the best car in the world for longer than others! lol