Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Project Dojo - the SaaS Product?

This site may earn commission on affiliate links.

Buckminster

Well-Known Member
Aug 29, 2018
10,331
51,400
UK
Interesting thread on Project Dojo:
Project Dojo

This discusses what I thought could be just the initial version of Dojo. It occurred to me that Elon could be thinking bigger (following the simple task of making Teslas autonomous...). Other possibilities:
  1. Google alternative
  2. Openai competitor (Elon removed himself for a reason?)
  3. Neuralink input / output
  4. Self learning capability
  5. Tesla insurance (see comment below from Fact Checking)
Could Project Dojo encompass 10,000 FSD computers - a net of neural nets? Could you throw Youtube at it and it learns how to speak 100 languages with little or no training?

And another interesting comment in another thread:
Neither Windows CE nor Symbian created a new market: they competed in the established phone market.

The iPhone and iOS created the new market of smartphones with touchscreens, and Apple has enjoyed the fruits of the resulting first mover advantage, all one trillion pieces of it. ;)

The new market of smartphones was inaccessible to Windows CE and Symbian, and by the time Microsoft and Nokia released their own smartphones it was too late: Apple captured 90% of the smartphone market profits for the next 10+ years.

So in that sense Windows CE and Symbian are like ICE car makers: they are established players in a dying market Tesla is not competing in.

Tesla is dominating the newly created EV market, where the late coming ICE makers have trouble competing.

To go back to the original argument:



Yes, it's the first product in a new high-tech market segment that gains the "first mover advantage", not the first technology.

This is why Microsoft won their PC monopolization war: they were businessmen first, technologists second.

This is why Apple eventually bested Microsoft: Steve Jobs was a product architect first, businessman second, technologist third.

This is why I think Tesla is going to prevail over the ICE industry, Elon is like Steve Jobs with empathy. "Product first" is written large in all companies of Elon, and a product by Elon always has to come with a top notch, first principles business plan.

I'm super excited about Tesla Insurance, not primarily because of the effects on Tesla's finances and SP (it will take years to take off), but because this is the first time since PayPal that Elon is launching a major high-tech financial product.

If the past is predictive, we might see something special - maybe not in the offering itself (which might look deceptively conservative, like a Model S), but under the hood. :D
 
FSD Dymaxion: [ FSD cluster computer system with Neural Net Machine Learning ]

Dymaxion
is a term that Buckminster Fuller associated with much of his work—prominently his Dymaxion house and Dymaxion car. Dymaxion, a portmanteau of the words dynamic, maximum, and tension; [1] sums up the goal of his study, "maximum gain of advantage from minimal energy input."[2]

inspired by @Buckminster :confused:
 
  • Love
Reactions: Buckminster
Weekend ATH cheers!

And with that, Elon tweeted about something I believe people are not paying enough attention to:


Here is my ATH hyper bull case:
  • Tesla is going to become the new Google+AWS for the era of software 2.0
What is means:
  • AWS generated 35B revenue last year and contributed more than half of AMZN’s profit, it’s a very lucrative business model.
  • I think Cloud computing for Software 2.0 (aka AI in hype word terminology) would be even bigger pretty soon.
  • Tesla would have opportunity to provide services on top of their physical world understanding, also able to provide computing 2.0 as service.
Here is why Tesla:
  • Cloud computing for AI is still an area everyone is fighting for, but no clear winner yet.
  • Tesla with FSD chip and now Dojo will be one of the best platform for inference and training for image/video understanding, aka physical world understanding.(THE best if you ask me)
  • Tesla would have millions of eyes on the streets all the time. With Tesla’s data source lead, they will become the leader in real world understanding. Google’s attempts for physical world understanding for example street view etc would look like toys in comparison.
  • Tesla with the vast amount of real world data, combined with battle tested hardware would be able to attract best talents to join (their mission statement helps too). This alone would be a very important factor for success.
I suspect when Tesla decided to design their own FSD chip, they already have this in mind. Marginal cost for producing chips is very low compared to R&D costs, once they have the capacity to build data centers with their own chip, the margin would be astonishing.

I assign this to a hyper bull case, not because I doubt Tesla’s capability of doing it, only because it seems to not align well with the current mission.

It might happen when Tesla expand their mission again though, let’s see.
 
Mr. Uujjj worked at a machine learning chip startup which was working on a very large computer for training neural nets. (This company has since emerged from stealth and got quite a lot of hype.) Elon Musk visited us as a prospective customer. Apparently Elon Musk was interested in buying us, but decided against it due to the price, and started an internal project at Tesla to build its own training machine.
 
Mr. Uujjj worked at a machine learning chip startup which was working on a very large computer for training neural nets. (This company has since emerged from stealth and got quite a lot of hype.) Elon Musk visited us as a prospective customer. Apparently Elon Musk was interested in buying us, but decided against it due to the price, and started an internal project at Tesla to build its own training machine.
So I saw the AI day presentation on Dojo. Pretty impressive. Though one thing I was struck by a similarity between Dojo and a system design presented in one of said startup's early slide decks, before they switched to what became the final design. I'm gonna have to ask some former coworkers about this 🤔
 
So I saw the AI day presentation on Dojo. Pretty impressive. Though one thing I was struck by a similarity between Dojo and a system design presented in one of said startup's early slide decks, before they switched to what became the final design. I'm gonna have to ask some former coworkers about this 🤔
Tesla has an "Invidia - chips - cluster" - which I suspect Tesla built. Invidia guy in an interview told this when asked about loss of chip sales to Tesla when Tesla designed the FSD computer. I don't know any details.
 
Friedburg estimates Google search costs $0.025 per search. ChatGPT is around $0.30 per search.

37 mins in:
Also, will get commoditised and therefore plenty of customers if Tesla is interested.

Following section is also pertinent - beating Google won't be easy obviously.
 
  • Informative
Reactions: UkNorthampton