Captkerosene
Member
Hard for Bot to hurt anyone from what I saw. You need to wheel it around in a wheel chair and all it does is wave at people.
I was hoping for a little more.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
I was hoping for a little more.
Krugerrand
Meow
It seems your task is to get invited next time so that someone is asking the right questions, eh?
Bet TSLA
Active Member
The hardware question that I thought would be most interesting to hear an answer to didn't get asked: what about skin? Do they see receptors for heat, cold, pressure, pain to be irrelevant for the early iterations? Surface flexibility? Just kind of curious what they're thinking. Hard problem.
Artful Dodger
"Neko no me"
Tesla Bot, AI Day, FSD, AI, AGI, and More with James Douma | Farzad Mesbahi (2022-08-19)
Get ready for Tesla Bot! (It's not what you think) | James Douma | TeslaHerbert (2022-08-16)
Cheers!
Xepa777
Banned
1) the progress tesla made in a year was amazing.
2) I told y’all here and Reddit not to expect crazy demos (reading predictions like jugglign or an audience member asking to pick a color from a set of items on the floor). You’re only disappointing yourselves (or getting your hopes up from hype influencers).
3) the actual robot demos were not “state of the art”. Again, I also pointed this out when referencing one of the major robotics professors at UCLA that works with/keeps in close contact with the Tesla team.
4) actual revolutionary robotics demos aren’t that important because Teslas strength is in mass profitable manufacturability and integration with software.
5) while the software today for the bot want impressive, the roadmap was (and details of which neural nets will actually be used with the bot).
Innovation disappoints in the short term and surprises in the long term. The bot has a very low bar to clear to be meaningfully profitable for tesla even with a limited set of use cases. It’s going to clear it.
2) I told y’all here and Reddit not to expect crazy demos (reading predictions like jugglign or an audience member asking to pick a color from a set of items on the floor). You’re only disappointing yourselves (or getting your hopes up from hype influencers).
3) the actual robot demos were not “state of the art”. Again, I also pointed this out when referencing one of the major robotics professors at UCLA that works with/keeps in close contact with the Tesla team.
4) actual revolutionary robotics demos aren’t that important because Teslas strength is in mass profitable manufacturability and integration with software.
5) while the software today for the bot want impressive, the roadmap was (and details of which neural nets will actually be used with the bot).
Innovation disappoints in the short term and surprises in the long term. The bot has a very low bar to clear to be meaningfully profitable for tesla even with a limited set of use cases. It’s going to clear it.
Artful Dodger
"Neko no me"
Robinhood buys Say Technologies for $140M to improve Shareholder-Company Relations | Techcrunch.com (Aug 10, 2021)
Haha, "payment for orderflow" indeed... when they have you LOGIN CREDENTIALS!
That's not how the Plaid API system works...Robinhood buys Say Technologies for $140M to improve Shareholder-Company Relations | Techcrunch.com (Aug 10, 2021)
Haha, "payment for orderflow" indeed... when they have you LOGIN CREDENTIALS!
Fidelity API & Data Solutions | Plaid
We got detail on motors and hands, but nothing on skin or haptics (or other senses). Statement about improving over time, but no plan.
Reasonable level of detail on this.
The training for tasks using a human to show the basic task and then software to compute variations was interesting.
Not really covered, it all seemed to be training on the servers - not learning in the bot.
Designed for manufacturing, they obviously have plans for manufacturing but are not sharing.
Pretty comprehensive update.
Lots of FSD training, if they mentioned HW4 I must have missed it.
Not mentioned.
Not mentioned.
Not mentioned.
DOJO as a service and Bot customisation mentioned, but no real detail.
So in the main this AI day 2 met my expectation.
What struck me was how much effort seems to be placed on performance, the number of times latency, compute times and performance improvements were mentioned was high. DOJO is all about how do we get performance in the rest of the system, not in the compute nodes (that are relatively easy).
dhanson865
Well-Known Member
Non issue if the chip is better, it's on the wrong platform. Dojo is replacing dozens of server racks with a few dojo tiles. The cost savings of the platform dwarf the chip efficiency difference you asked about.
Are you forgetting that a bunch of the people there were just random shareholders that were invited from the lottery?
To be honest, I fell asleep halfway during the presentation it was way over my head - that is why I cannot say how far they are ahead or not when compared to for instance Google or Boston Dynamic.
What I can say is that; whenever anyone says that Tesla is “just a car company” let them watch this presentation.
I also see that software is everything. If FSD is “solved” then - with few adaptations - Optimus is solved too. The hardware for Optimus would not be too difficult as far as I can see it.
What I can say is that; whenever anyone says that Tesla is “just a car company” let them watch this presentation.
I also see that software is everything. If FSD is “solved” then - with few adaptations - Optimus is solved too. The hardware for Optimus would not be too difficult as far as I can see it.
Artful Dodger
"Neko no me"
Tesla Shanghai was down for between 2-3 weeks for each production line upgrade in Q2. And we just found out the 'National Holiday' will be just a 2-day break for Giga Shanghai instead of 7 days in past years. That's ~30% right there. Now, add in the 22K/wk production upgrade...
Auto and Energy will turn out to just have been there to provide funds (and a foil in the case of FSD) to support AGI development. Sort of, but not quite, like a side hustle. (It’s similar to the likelihood that the money in Boring tunnels will be on the real estate/land use side rather than the digging of tunnels or the running of cars in them.)
Don’t feel bad if it doesn’t register yet—it won’t for most anyone else—though if you hold long enough it will. It seems that Elon himself is only realizing these things as he is going along. Fortunately, he has the bit between his teeth on FSD and Optimus. (Also, Elon mentioned ’emergent property’ a couple of times wrt intelligence so likely somebody has been reading Stephen Grossberg.)
The Future descendeth upon us with accelerating fury.
Last edited:
Did you guys watch the same thing I did? Optimus not only walked on stage, but turned to it's left, and then danced just like Elon! Granted, it was subtle, but the more I watch it the more impressed I am. That robot will be running the 660 next year, throwing strikes along the way... very impressive!
Well of the H100 gives a 6x gain vs a A100, then that exactly what Dojo is replacing so it's saving as much money as a H100 vs an A100.
It's kind of hard to compare the two because the compiler makes all the difference. I would have asked if Tesla has problems leveraging peak BF16 performance of an A100 vs Dojo. Also the H100 is on the newest Tsmc 4nm and it seems to be using 700w per gpu with 1petaflop of bf16 performance. Dojo is using 100kw per 54petaflop of performance. However is this even apples to apples? If so then Dojo is using 2x the power. Wish some people in the audience are tech journalists so they can ask some questions.
Artful Dodger
"Neko no me"
Artful Dodger
"Neko no me"
Dangerous, boring, or repetitive jobs. There's lots of those on Earth.
Hey, welcome to my life! :/
Similar threads
- Locked
- Locked
- Locked
