nothing. no link, just a guess. that's just how they communicate when key people leave to soften the blow.

jerome guillen went from president of automotive to the "trucking division" (does that even really exist?) early this year and then was gone a month ago.

FSD Beta 9 is finally here. It's robo taxi ready clearly, as there are only some near collisions and wrong lanes and other minor stuff in this short video. But the guys are really impressed how it turns in few corners, so there's that.

What's the latest official time frame on 'robo taxis'?

Who cares, stock moon, checkmate loser

You should avoid paying attention to noise. Important part of being good at investing.

When will the bears ITT admit that your thesis is wrong? Have any of you reviewed your flawed approach to understanding this business to figure out what you missed in order to avoid the same mistakes moving forward?

For bears avoiding self criticism, consider that if Tesla delivers FSD within the next decade, you will have been permanently wrong, you will die with Tesla >$1T market cap (minimum). This is of course not guaranteed and if they fail the stock probably gets cut in half.

It is amazing how close we are to being able to shut this thread down if Tesla executes. I wonder how many of you will have learned anything from this experience, let alone evolved to something greater than a fish in this game.

This thread is a support group for losers, its been allot of fun to observe.

Tesla will never deliver FSD to any vehicle they have already sold. I will admit being wrong about the market's appetite for risky car companies. But I am 100% correct that Tesla will never deliver FSD (for current vehicles). Just a matter of waiting now.

Never is a very long time.

TSLA cars actually already can drive themselves for a decent part of a trip. Does anyone disagree with this?

Why do bears think that going from the current state to a 99.9% FSD is impossible. What is the fundamental argument for why TSLA can get to this level, but never to a higher level?

I'm genuinely curious. I think AI is extremely effective in bounded games/areas like poker/chess and way harder to solve in RL settings. However, TSLA seems to be improving yoy. So what's exactly the bottleneck here?

Because even if Tesla had 99.9% FSD, that's not even close to FSD. Those edge cases matter a lot, human driver dies every 0.00000125% of the miles they drive and human drivers really suck. You can't really trust a FSD that is like 5 magnitudes worse than a human driver.

Plus they have really shown that they do not actually have a way to create one. With constant promises, really bad under delivery, firing the people associated with FSD, and now lately they changed from radar+cameras to just cameras. They can market those actions any way they want, but in reality they only tell us that their method before didn't work and they don't think they can make them to work, so they're starting fron scratch.

So why will they never solve the edge cases? Why did they get to this point, and won't go much further? Is it data, computing, algo? What's lacking in your opinion to solve this problem? Can nobody solve it, or will other companies be able to?

Also a small sidenote, but a 99.9% FSD to me doesn't mean that a person dies 0,01% of the time, it just means there are still certain edge cases TSLA doesn't handle well, but overall it could be a much better driver than humans at that point.

The edge cases are the problem with making a FSD. Everything else before that is the easy part that has been solved multiple ways a long time ago. Them going back to drawing board and making a completely different system just shows they couldn't fix the old one, which is not surprising at all.
These differences do matter a lot, but I think that's just bad for Tesla. Driving yourself, you know when situation gets dangerous and you concentrate and try to get out of it. With a shitty FSD that is 99.999% correct you should be absolutely shitting your pants and ready to act every moment you're using it.

I appreciate the answers, but it still doesn't explain to me what so different about edge cases vs the easy part of solving this problem. Edge cases make it sound like there's a lack of data. But then there's a world where TSLA has 5M+ cars on the road and years of data collection where they actually do get enough data even for the edge cases in 5-10y?

Because how these machine learning systems work, they're basically black boxes where you feed data and get an answer. Usually it's not a problem when you get a wrong one, but in a car it can be an accident. When your ML doesn't give you the correct answers, you turn the knobs and tune the input and hope you can get better ones. Which is ok if you're aiming for 90% accuracy, but not going to help when you aim for 100 or even 99%. Limits of what benefit an unlimited amount of data can give you fizz out way way way before a reasonably safe FSD%, Machine learning systems are just way too limited currently to recognize these well and all the other methods are not even close to handling the unlimited edge cases of real world.

I'm not so sure these problems will be even solved this decade, because the last 0.001% or whatever is just from so different world of problems than the first part, but it surely doesn't look like Tesla will be the one to do it if it happens.

Take GPT-3 as an example, which probably is one of the most complicated AIs today. It took $12 million computing power just to train it, with simple text, no images or video.

It's ok if you can refresh it 10 times and choose the best result. Using similar AI for your autopilot or at least for most data feeds for your autopilot code from much, much, much more complicated real world video? Nah, not even close.

Tesla has been promising (and selling) FSD since 2016. When do you stop giving them the benefit of the doubt and admit it's a vaporware scam?

Real autonomy is many orders of magnitude harder than a level 2 driver assist (which plenty of other car companies have as well). Tesla just happens to be the only one desperate and dishonest enough to pretend that actual full self-driving is just around the corner.

Have you looked at what competitors such as Cruise and Waymo are doing? They have an actual viable strategy for self-driving, vs. Tesla's pipedream of taking a bunch of shitty camera data and throwing it at a neural net. And they're way ahead of Tesla on every other dimension too: more robust hardware, diversity of sensors, data collection, HD mapping, testing, simulations, R&D budget, programming talent, etc.

Waymo has been testing in Phoenix/Chandler for 3 years, and has been "fully operational" (in one city with safety drivers) for about 1 year. They still can't make unprotected lefts and they still can't merge onto the freeway without the AI crapping the bed.

Waymo also has the problem of setting up not only service infrastructure for its autonomous taxis, but runs into the question of whether people will even buy it: arstechnica.

I've watched a couple interviews with George Hotz from comma.ai. He seems to be thinking that TSLA is on the right path and will probably be first but that they're not that far behind. He wasn't very optimistic about Cruise/Waymo.

FSD obv isn't just around the corner. Musk is dishonest and has no morals, however eventually things do get build and shipped. From what I've read and seen there seems to be quite a difference between AP 2016+ (I dont know when they first started using AP) compared to now. Vaporware to me means something that actually does not work at all. However, these cars are actually out there on the roads, in real life.

^This is why you tell big lies. Idiots will partially credit them and hold the bright shining big lie as a future hope.

Musk has level 2 driver assist with a few level 3 features. He has zero FSD and is not on the path to FSD in the next 5 years. It's all lies and fraud.

They might not ever attain FSD, but their government subsidy game is on-point:

https://www.thedrive.com/news/40975/...rs-115-million

That guy is the Freddy Adu of self-driving. I remember when he shot to fame as this great prodigy who was going to solve self-driving, doing it with spare parts like Tony Stark. I think Elon wanted to hire him, but he turned him down and caused some beef. What has comma done since its inception a few years ago? Sold a few DIY kits to Honda owners so people can install their own adaptive cruise control and feel like they're hackers?

it's driverless but in a city with almost no rain and no snow, all streets are a grid with no hills etc.

Oh look, another non-fraud company lightyears ahead of Tesla. Mobileye self-driving in NYC (vision only):


vs Tesla shitshow:
https://www.businessinsider.com/tesl...litches-2021-7

Tesla can't even handle parking lots and driveways.

Waymo in a parking lot. You can see the other cars driving around it and passing it in the parking lot because the Waymo isn't moving.

Sorry, I should have been more precise--Tesla can't handle empty parking lots and driveways. But anyway, what's your point? Waymo struggles in a busy parking lot, therefore Tesla full self-driving isn't vaporware fraud?