You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
Hopefully one day in the not too distant future she will be wistful that she didn’t buy-in when she had the chance..
Media's agenda and modus operandi is very simple: egg the regulators, shame them, goad them so that they will come out and lash at Musk and cut him to size. They have tried everything : writing FUD, total lies, throwing shade that Musk's personal character, scaring people with exaggerated Tesla fire stories. Nothing seems to have an impact. Now they are all angry at the regulators and shouting them: DO SOMETHING, will ya. !!FUD from NPR written by David Gura on June 4. 2022-
Can the SEC stand up to the richest man on the planet?
Well, it's a nice theory. But in the real world, every iteration means that some tests that used to fail now pass, and some tests that used to pass now fail. Not all unit tests have the same importance, so any score is really determined by how the relevant humans value the test cases. A simple metric of "no regressions allowed" would probably mean never releasing a new version.Nice FSD drive:
My take is that the big improvements are mainly that they have improved their cost function in their control system. It now rewards having better visibility by creeping and angling the car and also penalizes being stuck in dangerous situations, ie better to have some ”uncomfortable” jerk than driving slowly through the dangerous intersection.
View attachment 812846
In the next version 10.13 they will again improve the cost function to penalize having the car slightly outside of the median to reward angling the car so it fits in the median even if this means a more lateral jerk.
So first they will add this behavior. It will not be perfect in the first iteration, but it will be part of the score. Then they will take another iteration to gather data, tune it and improve its performance to get closer to perfection.
This is why I believe in Tesla. Instead of figuring out what to do they first figure out what they want to accomplish and then let an algorithm optimize how to do it. If they are not happy with what they algorithm suggests, they improve the cost function to better reflect the desired behavior. Same with their agile development, they have a cost score ie ”make as many safe cars at a profit as possible”. If some engineers comes up with some idea ”let’s do X instead of Y” they see if this improves their score or not. If it does they implement it, or if something seems off, they instead change their value function.
Like Karpathy shows they have an increasing number of ”unit tests” for different scenarios. Whenever they recompile their code they see if they get a higher score, ie if their new release passes more unit tests.
View attachment 812845
When their customers are out driving and the system fails, they add the situation to their unit tests. Then they let their engineers and operation vacation try to add more data, improve the neural network, tune control paramaters etc freely, as long as they improve how many of their unit tests they pass and/or their cost function. That’s how they are making progress, by improving their cost function, by adding unit tests and by improving their score on these tests.
In a truly automated machine learning system, regressions are mostly not manually handled by the team except under obviously serious circumstances and edge cases where the machine learning is simply not going to be able to experience it often enough to learn to do it right. Every FSD beta so far has had large advancements combined with regressions and that will continue and that should continue. It's how real world learning works, and if the FSD beta testing driver is paying attention properly and ready to take over if needed, then it is working as intended.Well, it's a nice theory. But in the real world, every iteration means that some tests that used to fail now pass, and some tests that used to pass now fail. Not all unit tests have the same importance, so any score is really determined by how the relevant humans value the test cases. A simple metric of "no regressions allowed" would probably mean never releasing a new version.
So who determines which unit tests are "must pass" and which are fine to be part of the "one step back" cohort?
Me, I remember one similar situation at Apple where every change the team tried resulted in unacceptable regressions. It was... stressful.
It’s fun to imagine the firestorm of media shock and dismay if Tesla started day trading its own stock!Tesla should have a share buyback program so they could take advantage of false negative media reports, especially Kolodny events. Tesla has no responsibility to comment on media speculation, so it wouldn't be a violation to say nothing and buy back shares.
I get your point, but I’m going to venture a bit further out on the limb. The pattern I think I’ve been seeing, is that positive news can drive an upswing, but MMs can fight it if the volume is low. Negative news gets amplified, and then any macro trends have a pile-on effect. That nice limb might get sawed off by Monday morning.
On a somewhat related note, I got an email from my sister referring me to a NYT OpEd entitled, “When Elon Musk Dreams, His Employees Have Nightmares.” I had seen the headline as I have a digital subscription to it, but didn’t waste my time reading it, nor did I when she cited it. My sister was sort of poking me. She and her husband are Porsche fans. I sent her back the following note:
‘Thanks. Yes I saw this article. He was pilloried in the press after he first said all workers needed to work 40 hrs in the office, and then said he was reducing staff by 10%. He later clarified that to be just office workers, and that they are increasing factory and service workers. It caused the stock to drop a good bit.
Here’s the thing, he is pushing the envelope. For those who want to work at a place where there is constant innovation, challenge, and empowerment to achieve - Tesla is it. And more do than don’t want to be there. It isn’t-a great place for softer skills most likely. And it’s probably not a place you’d spend a career at.
As an investor, I’ve learned to filter the things he says and does through a net that drops out impacts to core business. I’ll admit it is sometimes difficult. But so far when I do this, what I see is a company that is growing dramatically, by constantly innovating and adapting to near-term, medium-term, and long-term constraints, as it maniacally focuses on achieving its objectives. They are not for faint of heart. I periodically check in with people whose financial knowledge and wisdom I respect just to make sure I’ve not got rose colored glasses on. And I read an investors’ forum daily. Staying the course.’
Of course the investors’ forum I referred to is TMC/this thread.
Hopefully one day in the not too distant future she will be wistful that she didn’t buy-in when she had the chance..
13:30 here:Well, it's a nice theory. But in the real world, every iteration means that some tests that used to fail now pass, and some tests that used to pass now fail. Not all unit tests have the same importance, so any score is really determined by how the relevant humans value the test cases. A simple metric of "no regressions allowed" would probably mean never releasing a new version.
So who determines which unit tests are "must pass" and which are fine to be part of the "one step back" cohort?
Me, I remember one similar situation at Apple where every change the team tried resulted in unacceptable regressions. It was... stressful.
This is a Tesla warehouse 2 km away from the Shanghai factory, "the supplier parts are stored here, and then distributed according to the needs of the production line, through this Shanghai lockout, Tesla will increase their inventory to prepare for emergency needs."
Also in this twitter thread :This is great news.
I just got invited to FSD Beta and it has been closed for months.
I was in the program on my model three, but lost out when I bought when I traded in for the model S.
FYI I have a 100 Safety Score.
We are close to the point where offering something for EU regulators to review makes sense
Interested to hear from 2018 model 3 new invites. Are you recently (June 2022) invited?This is great news.
I just got invited to FSD Beta and it has been closed for months.
I was in the program on my model three, but lost out when I bought when I traded in for the model S.
FYI I have a 100 Safety Score.
Hmmm. Wondering if expanding FSD Autosteer on City Streets to 100k cars will release a portion of the Deferred FSD Revenue to Q2 Earnings?
Per the Q1 10Q, Tesla expects to earn about $1.6B in the next 12 months from revenue that had been deferred. I wonder how much of this $1.6B will show up in Q2? if any?
View attachment 812796
Turns out, I can confirm. With 99 safety score, it just installed and I took a confirmation drive.Interested to hear from 2018 model 3 new invites. Are you recently (June 2022) invited?