Elon: FSD Beta tweets

[Knightshade]

Who cares. FSD beta is what I care about, and it’s got a long way to go. It doesn’t matter who has the best sensors or the best neural net. It matters who is best at exploiting what they do have. Anything else can be bought.

that's...sort of the point of Dojo.

Nobody had anything Tesla could buy that would do what they want fast enough for training massive video data sets. So they designed their own.

Elon even mentioned that a measure of Dojos success once it's built and in service will be its ability to exceed the capabilities of the best buy-someone-elses-parts solution they already have right now (the giant A100 supercomputing cluster)

 

[qdeathstar]

that's...sort of the point of Dojo.

Nobody had anything Tesla could buy that would do what they want fast enough for training massive video data sets. So they designed their own.

Elon even mentioned that a measure of Dojos success once it's built and in service will be its ability to exceed the capabilities of the best buy-someone-elses-parts solution they already have right now (the giant A100 supercomputing cluster)

That’s what they think, of course, but Elon, like mankind, has a track record of being wrong on some things and right on somethings. Just because he believes it doesn’t mean it’s true.

Autonomous driving was a solved problem long before DOJO according to him…

 

[EVNow]

Here is something novel for this thread - an Elon tweet.

https://twitter.com/x/status/1471215375215902729

 

[EVNow]

And ...

https://twitter.com/x/status/1471154807557836801

 

[EVNow]

To go with the earlier ....

https://twitter.com/x/status/1470892106776514566

 

[nvx1977]

From the FT interview:

On the challenge of autonomous driving, meanwhile, he gets close to conceding that some of his own earlier confidence was misplaced. “I didn’t think we would have to solve a significant part of artificial intelligence to make it work”. But he adds that he is now “99.9 per cent — round it up to 100 per cent — confident full self driving will work, it’s just a question of when”.

Good to see him change his tune. He has no idea when FSD will happen, but he thinks it def will happen. Of course. If you don't believe that, you won't have the drive to tackle the problem.

 

[impastu]

"he is now “99.9 per cent — round it up to 100 per cent — confident full self driving will work, it’s just a question of when”.

And as macro economists like to say, "in the long run, we're all dead"

 

[Bladerskb]

Yeah you've got a guy who has no idea WTF he's talking about (again) being caught making that obvious (again) and now trying to move goalposts about what "specialized" and "generalized" hardware actually mean to make it less obvious (again).

At least he's consistent, even if it's consistently wrong

BTW, excellent additional example of the actual meanings citing DSPs. They can do LOTS of jobs generally, but you can still absolutely build ones that do specific things better than others (and most do so)....

Post your credential and i will post mine, lets see who actually has experience in ML and who has no clue what they are talking about and just regurgitating whatever elon says.

 

[Bladerskb]

Are Dojo's 8 and 16 bit floating point numbers specialized?

Tesla releases Dojo whitepaper, Elon Musk teases as 'more important than it may seem'

Tesla has released a new whitepaper regarding a new standard for its Dojo supercomputing platform. CEO Elon Musk teased the...

electrek.co

Quote:
This standard specifies Tesla arithmetic formats and methods for the new 8-bit and 16-bit binary floating-point arithmetic in computer programming environments for deep learning neural network training.

When I say something isn't specialized, I don't mean its not specialized for ML.
NN processor perform matrix/multiply operations, that's it.

 

[Knightshade]

When I say something isn't specialized, I don't mean its not specialized

I think this sums up your posts extremely well

 

[Bladerskb]

I'm not sure that is the case at all. NN hardware is advancing in terms of performance and complexity, but the underlying architecture has been evolving relatively slowly.

Actually its not, anyone who has been following the ML community would tell you that, even people who tell you its recently slowed down.
But in the context of chip development, we are moving in light years speed. Remember it takes up to 3 years for a new chip to be deployed commercially. In 3 years the NN architecture landscape is completely different.

The creator of Intel's Nervana said the same thing.

No way ... things can be "specialized" in the sense they are aimed at specific vertical applications but are still flexible enough to adopt to differing needs .. for example a DSP for audio processing etc.

I never said they couldn't be. Their performance however suffers when you try to use it for something else, thats why chip designers who are creating NN accelerators don't do it.

 

[Bladerskb]

Google v4 may well be faster at OTHER tasks of course. But why would Tesla care since those aren't the ones they need to be faster?

that's...sort of the point of Dojo.

Nobody had anything Tesla could buy that would do what they want fast enough for training massive video data sets. So they designed their own.

Elon even mentioned that a measure of Dojos success once it's built and in service will be its ability to exceed the capabilities of the best buy-someone-elses-parts solution they already have right now (the giant A100 supercomputing cluster)

This is nonsense and just regurgitating rubbish.

 

[impastu]

Actually its not, anyone who has been following the ML community would tell you that, even people who tell you its recently slowed down.
But in the context of chip development, we are moving in light years speed. Remember it takes up to 3 years for a new chip to be deployed commercially. In 3 years the NN architecture landscape is completely different.

The creator of Intel's Nervana said the same thing.



I never said they couldn't be. Their performance however suffers when you try to use it for something else, thats why chip designers who are creating NN accelerators don't do it.

dude, you need a day job...

 

[Knightshade]

This is nonsense and just regurgitating rubbish.

Just because you don't understand it doesn't make it untrue

 

[Bladerskb]

dude, you need a day job...

You should listen to a Chip Architect rather than cheerleaders on a tesla forum with no ML or chip experience.

Naveen Rao - Chip architect, co-founder of Intel Nervana on Dojo

"The NN approaches change every year so there's a pretty big shift on how things are done. They were smart enough to build this in a general way. From what i can tell it looks like relatively standard generalized AI compute. Meaning that it can handle any variety of different type of NN approaches. Which is smart. You don't want to build a piece of hardware that's not going to be ready to be usable for 2 years and then in 2 years all of a sudden everything shifted and it doesn't work. You can't make it so specific. They did it the smart way."

 

[Bladerskb]

From the FT interview:



Good to see him change his tune. He has no idea when FSD will happen, but he thinks it def will happen. Of course. If you don't believe that, you won't have the drive to tackle the problem.

Wait till January when he renews his yearly prediction and guarantees.

 

[Matias]

"he is now “99.9 per cent — round it up to 100 per cent — confident full self driving will work, it’s just a question of when”.

And as macro economists like to say, "in the long run, we're all dead"

Is there some rough estimate, like will it be ready before Sun turns into red giant?

 

[DarkandStormy]

Tesla tells customers it will not honor retroactive Full Self-Driving pricing

Tesla is communicating to customers who have ordered now-discontinued variants of vehicles that it will not honor retroactive pricing of the Full Self-Driving suite to when they reserved their vehicle. In February, Teslarati reported that Tesla would not produce the Model Y Long Range...

www.teslarati.com

I'm sure there's some tweet about locking in FSD price at purchase, right?

 

[EVNow]

Tesla tells customers it will not honor retroactive Full Self-Driving pricing

Tesla is communicating to customers who have ordered now-discontinued variants of vehicles that it will not honor retroactive pricing of the Full Self-Driving suite to when they reserved their vehicle. In February, Teslarati reported that Tesla would not produce the Model Y Long Range...

www.teslarati.com

I'm sure there's some tweet about locking in FSD price at purchase, right?

I think the issue is if the model you ordered is no longer being produced and you are being asked to choose a new model.

Since I didn't want to buy this year, my Model Y order was cancelled by Tesla sometime back. Apparently they didn't cancel everyone's ...

 

[Knightshade]

I think the issue is if the model you ordered is no longer being produced and you are being asked to choose a new model.

Since I didn't want to buy this year, my Model Y order was cancelled by Tesla sometime back. Apparently they didn't cancel everyone's ...

AFAIK the cancelled order thing was folks who had been delaying delivery of existing ready-to-be-delivered configs for many months. Tesla finally said if you don't want the car after X many months of it being offered to you we're killing the order.


What's interesting is- Y buyers who ordered a model that never got made had no promise about FSD pricing for a different config.

But Cybertruck preorders WERE told they were locking in price by adding FSD at reservation time.

So it'll be interesting for the 1 and 3 motor configs that likely won't ever exist- if they'll honor that when they switch to 2/4 motor configs.