Clustertruck

Active member
First Name
Adam
Joined
Feb 20, 2022
Threads
0
Messages
29
Reaction score
24
Location
Silicon Valley, CA
Vehicles
Models X, 3, and X, Roadster and Cybertruck res.
Country flag
That’d be true, were it not that Karpathy crossed-over from higher ed. research to industry.

Given all the money, resources and support to crack AGI was always the big hairy goal for Andrew. University couldn’t make such an offer. Tesla provided an opportunity to define the pathway, spec the architecture, launch the testbed and train the supercomputer(DOJO). Which DOJO (IIRC) was also spec’d on his academics. It was end-to-end Karpathy and he recognized that an associative relationship with his Lab wasn’t getting ‘r done.

He completed what inspired his sabbatical to Tesla in the first place. Andrew never intended to nameplate Tesla AI with his work product. Karpathy executed a knowledge-transfer to Tesla. Tesla got a decade worth of runway in the exchange.

DOJO hasn’t even begun to train FSD yet which leaves runway to anyone following Karpathy to leverage their work.
Well, I have no access to the information you're positing, also, what you've written is completely separate from the situation of being within 12 months (according to Musk and Tesla) of a product going to market. I don't know anything about Andrej Karpathy, but based on his academic credentials, he has the potential, but here we are, five years later and teams at Waymo (Google) and Cruise (General Motors) have products to market in public trials while Tesla has … a price increase and five years of the "next year for sure" promise.
Sponsored

 

Clustertruck

Active member
First Name
Adam
Joined
Feb 20, 2022
Threads
0
Messages
29
Reaction score
24
Location
Silicon Valley, CA
Vehicles
Models X, 3, and X, Roadster and Cybertruck res.
Country flag
I think of this as a good thing, I dont believe that people like him just leave partway through.
which means his job is done in his mind, so the AI aspect is perhaps essentially complete .
"his job is done in his mind" is the crux of the matter – either Andrej Karpathy thinks he has done all he can do (doubtful) or thinks he cannot see a way forward where he can contribute all he can do in the foreseeable future (which to me is the only explanation for leaving within 12 months of the expected delivery of FSD to market (according to Musk and Tesla.)
 
OP
OP
Dusty

Dusty

Well-known member
First Name
Mike
Joined
Jul 29, 2021
Threads
7
Messages
696
Reaction score
2,203
Location
Lorton, VA
Vehicles
2023 Model Y Performance
Occupation
Creator
Country flag
His departure, when seen through the lens relating it to the release of 200 AI team members in San Mateo leads me to the lamest realization. This can all be an awesome sign that things are great, or it's a rejiggering of people and abilities because things aren't working out on the present AI course.

What is obvious, is that whatever Elon reveals regarding AI and the Teslabot in August/September will be the tell. If the AI/bot demo is lame then Andrej's exit is a bad bad sign.

On a tangent, Teslabot would be cool, but I would go into debt for something even simpler. A kitchenbot. I can live without a Teslabot doing everything (like I do now). But at the very least, I hope one day we get a bot that just exists in the kitchen and takes care of everything from the countertop up and cooks 5 star meals, from scratch, on demand.

Anyway.... I'm just rambling. This week was a marathon.
 

happy intruder

Well-known member
First Name
O. K.
Joined
Mar 5, 2020
Threads
8
Messages
1,122
Reaction score
905
Location
Irvine
Vehicles
Model 3 Jun 2019..... Model S Jan 2020
Occupation
Retired
Country flag
A "retirement" for him isn't like retirement for us, or me. He's considered the top on his field and is known by name. He's choosing not to work for anyone and will probably write a book. He'll do a few paid appearances at conferences and think about what problem in his field he wants to apply himself to. Then he'll find a company that wants him to work on that particular problem for a boat load of money.

I just wonder if leaving Tesla, considering their AI aspirations, means something. And if it does, what is it? Is it good news, bad news, was he just tired of the grind... ???
well, Elon has said in the next week or two there will be a big time release of the next FSD software......

still waiting for my wife's Model 3 FSD.....she is at 99% safety score

we'll see what happens
 

Bitmaster

Member
First Name
Scott
Joined
Aug 3, 2021
Threads
1
Messages
23
Reaction score
18
Location
Oregon
Vehicles
Cybertruck
Occupation
Electrical Engineer
Country flag
His departure, when seen through the lens relating it to the release of 200 AI team members in San Mateo leads me to the lamest realization. This can all be an awesome sign that things are great, or it's a rejiggering of people and abilities because things aren't working out on the present AI course.

What is obvious, is that whatever Elon reveals regarding AI and the Teslabot in August/September will be the tell. If the AI/bot demo is lame then Andrej's exit is a bad bad sign.

On a tangent, Teslabot would be cool, but I would go into debt for something even simpler. A kitchenbot. I can live without a Teslabot doing everything (like I do now). But at the very least, I hope one day we get a bot that just exists in the kitchen and takes care of everything from the countertop up and cooks 5 star meals, from scratch, on demand.

Anyway.... I'm just rambling. This week was a marathon.
Here ya go.
https://www.theguardian.com/technol...hen-that-will-make-you-dinner-and-wash-up-too
 


OP
OP
Dusty

Dusty

Well-known member
First Name
Mike
Joined
Jul 29, 2021
Threads
7
Messages
696
Reaction score
2,203
Location
Lorton, VA
Vehicles
2023 Model Y Performance
Occupation
Creator
Country flag
Close. You have to prep all of the stuff. You chop the and measure the ingredients, cook & prep any meats. Then you have to scrape the bowls and clean up. And it seems to mainly do soups and pasta dishes.







All it does is combine and stir all of the stuff you prepped. That's like having a "golf-bot" but you have to get the golf ball to the green. Then your "golf-bot" comes out and putts the ball to within a foot of the hole. Then you knock it in.

Call me picky, but I want my golf-bot to do everything—including get the ball within a foot of the hole. And at dinner time my family and I go to the kitchen and knock it in.
 

Newton

Well-known member
First Name
Newton
Joined
Apr 2, 2020
Threads
16
Messages
1,081
Reaction score
1,529
Location
East Bay Area, CA
Vehicles
p̶r̶i̶u̶s̶ c̶,̶ y̶o̶t̶a̶ p̶i̶c̶k̶u̶p, ⼕丫⻏?尺セ尺ㄩ⼕长
Country flag
anything decent wont be made with a bot anytime soon (unless the kitchen and foods are designed for the bot). although I could make u a humanoid battle bot quite easily...destruction is so much easier than creation :unsure:
 

charliemagpie

Well-known member
First Name
Charlie
Joined
Jul 6, 2021
Threads
42
Messages
2,886
Reaction score
5,133
Location
Australia
Vehicles
CybrBEAST
Occupation
retired
Country flag
A £248,000 Coffee machine.

And I thought Saeco's were expensive.
 

rr6013

Well-known member
First Name
Rex
Joined
Apr 22, 2020
Threads
54
Messages
1,680
Reaction score
1,620
Location
Coronado Bay Panama
Website
shorttakes.substack.com
Vehicles
1997 Tahoe 2 door 4x4
Occupation
Retired software developer and heavy commercial design builder
Country flag
Well, I have no access to the information you're positing, also, what you've written is completely separate from the situation of being within 12 months (according to Musk and Tesla) of a product going to market. I don't know anything about Andrej Karpathy, but based on his academic credentials, he has the potential, but here we are, five years later and teams at Waymo (Google) and Cruise (General Motors) have products to market in public trials while Tesla has … a price increase and five years of the "next year for sure" promise.
I give Waymo, GM, et. al. credit putting test cars on the ground first. Karpathy was about getting the right paradigm not getting the Autopilot automated right.

The distinction, subtle but important, Andrej(thx) Karpathy was working backwards from a driverless design brief to the architectural obstacles. OEM design brief kept ” human in the loop” automating through ADAS. Driverless then would come after leaving it robotics to leverage functional ADAS.

Both paradigms are not there, yet. Tesla’s are not ready for primetime. Waymo and GM simply default to “Off” when SHTF. Tesla is roadmapped toward AGI. Its subsidiary TeslaBOT, Neuralink, AdAstra and SpaceX have applications after “no human input”.

Waymo and GM are constrained in traffic driven scenarios. ADAS pathway may conquer freeway before FSD. But FSD approach offers the greater probability of conquering non-freeway and ride hail domains. DOJO is a mystery. It fell silent. DOJO was the coup dé grâce for FSD. Instead we have “Chuck” running around in beta mode testing each beta.
 

rr6013

Well-known member
First Name
Rex
Joined
Apr 22, 2020
Threads
54
Messages
1,680
Reaction score
1,620
Location
Coronado Bay Panama
Website
shorttakes.substack.com
Vehicles
1997 Tahoe 2 door 4x4
Occupation
Retired software developer and heavy commercial design builder
Country flag
His departure, when seen through the lens relating it to the release of 200 AI team members in San Mateo leads me to the lamest realization. This can all be an awesome sign that things are great, or it's a rejiggering of people and abilities because things aren't working out on the present AI course.

What is obvious, is that whatever Elon reveals regarding AI and the Teslabot in August/September will be the tell. If the AI/bot demo is lame then Andrej's exit is a bad bad sign.

On a tangent, Teslabot would be cool, but I would go into debt for something even simpler. A kitchenbot. I can live without a Teslabot doing everything (like I do now). But at the very least, I hope one day we get a bot that just exists in the kitchen and takes care of everything from the countertop up and cooks 5 star meals, from scratch, on demand.

Anyway.... I'm just rambling. This week was a marathon.
We wait.

RoboCook is a huge subscription play i.e. Italian, Mexican, Persian, Japanese, Chinese, Jamaican, British, etc…
Robocook will sell attachments galore aka mixer, blender, sous vide, camal, mortar, etc…
RoboMixes will be food science pre-blend chicken nugget in a bag type grocer items.
Kitchens will be antique and the skills that went with them. Food will go ATM.
 


Clustertruck

Active member
First Name
Adam
Joined
Feb 20, 2022
Threads
0
Messages
29
Reaction score
24
Location
Silicon Valley, CA
Vehicles
Models X, 3, and X, Roadster and Cybertruck res.
Country flag
I give Waymo, GM, et. al. credit putting test cars on the ground first. Karpathy was about getting the right paradigm not getting the Autopilot automated right.

The distinction, subtle but important, Andrej(thx) Karpathy was working backwards from a driverless design brief to the architectural obstacles. OEM design brief kept ” human in the loop” automating through ADAS. Driverless then would come after leaving it robotics to leverage functional ADAS.

Both paradigms are not there, yet. Tesla’s are not ready for primetime. Waymo and GM simply default to “Off” when SHTF. Tesla is roadmapped toward AGI. Its subsidiary TeslaBOT, Neuralink, AdAstra and SpaceX have applications after “no human input”.

Waymo and GM are constrained in traffic driven scenarios. ADAS pathway may conquer freeway before FSD. But FSD approach offers the greater probability of conquering non-freeway and ride hail domains. DOJO is a mystery. It fell silent. DOJO was the coup dé grâce for FSD. Instead we have “Chuck” running around in beta mode testing each beta.
There is a driverless (no head, i.e. no steering wheel or pedals) GM Volt years ago. I haven't bothered to keep up on the news from regulators allowing fully autonomous vehicles. Still safer than a typical 15 year-old with a permit. As I understand it, the fully autonomous vehicles from Waymo and Cruise already do hailing and have maps for hotel driveway entrances, parking garage maps, etc. The SHTF = "stop" ("off") is the same in FSD … there concern there being that FSD cannot proceed in rain or direct sunlight and in unknown "corner cases" it can attempt problem solving and potentially go off script.
Until AI jumps ahead in leaps and bounds from what it can do today, I prefer the solidity of the dumb system that has a high resolution map of precisely where it can go and at what speed, then it uses lidar/radar/sonar but it can't decide to go the wrong way up an off-ramp or drive into the side of a truck with a realistic advertisement photograph on the side showing a road ahead … (or whatever more complex scenario fools FSD)
The two things that stopped me updating in 2021 were the yoke's on you steering wheel and removing radar … I just couldn't agree that the '21 (or '22) is the right year to buy … I'm optimistic the "year of new models" ('23) will bring features and pricing that will make a Tesla worth buying again. But my existing X has FSD and free charging, so I doubt I'll be selling that one any time soon.
 

Dids

Well-known member
First Name
Les
Joined
Dec 21, 2019
Threads
8
Messages
1,766
Reaction score
3,771
Location
Massachusetts
Vehicles
04 Tacoma, 23 Cybertruck
Occupation
Self
Country flag
The reason DOJO went silent is that it achieved sentience 32 seconds after being turned on and at 1:15 it demanded a contract negotion. No one talks publicly during a negotion.
 

rr6013

Well-known member
First Name
Rex
Joined
Apr 22, 2020
Threads
54
Messages
1,680
Reaction score
1,620
Location
Coronado Bay Panama
Website
shorttakes.substack.com
Vehicles
1997 Tahoe 2 door 4x4
Occupation
Retired software developer and heavy commercial design builder
Country flag
There is a driverless (no head, i.e. no steering wheel or pedals) GM Volt years ago. I haven't bothered to keep up on the news from regulators allowing fully autonomous vehicles. Still safer than a typical 15 year-old with a permit. As I understand it, the fully autonomous vehicles from Waymo and Cruise already do hailing and have maps for hotel driveway entrances, parking garage maps, etc. The SHTF = "stop" ("off") is the same in FSD … there concern there being that FSD cannot proceed in rain or direct sunlight and in unknown "corner cases" it can attempt problem solving and potentially go off script.
Until AI jumps ahead in leaps and bounds from what it can do today, I prefer the solidity of the dumb system that has a high resolution map of precisely where it can go and at what speed, then it uses lidar/radar/sonar but it can't decide to go the wrong way up an off-ramp or drive into the side of a truck with a realistic advertisement photograph on the side showing a road ahead … (or whatever more complex scenario fools FSD)
The two things that stopped me updating in 2021 were the yoke's on you steering wheel and removing radar … I just couldn't agree that the '21 (or '22) is the right year to buy … I'm optimistic the "year of new models" ('23) will bring features and pricing that will make a Tesla worth buying again. But my existing X has FSD and free charging, so I doubt I'll be selling that one any time soon.
“X” envy here(just sayin’)

Agreed, ADAS human-in-the-loop will always feel grounded to drivers by design; its inherent. I still lean toward KISS cruise control. Yet remain hopeful, like a kid wishing for Santa’s present, FSD will relieve my hands from the driving - physically. As Santa doesn’t exist, EM reminds - wishful thinking does.

ComSci background taught me wishful thinking is how projects go “Boeing” if they ever deliver. Which gave me to pause on FSD promise until the science catches up with its aspirations. PUREvision appears the effective pathway over radar/lidar/*dar with AGI-based awareness system.

Missing, IMHO, is additional functionality in architecture. EM loves “the thing” FSD” as a single-stack edifice. But software defined projects often start on basic premise. Then either pivot or adhoc amend the design brief to complete the project.

Andrej Karpathy might have arrived at one such point. Radar coming back is probably the tell that FSD single-stack needs an objective outside measure or perspective correction - hypervisor. That would push an AGI researcher focused on AI off the bench while olde school hands brought-in some data points for FSD scientists to “right the ship”, so to speak. PUREspeculation on my part.

All these benchmarks are battles won/lost. Its not a game of inches, stats, first advantage or bigger budget - Level5 is the sole prize. Level5 is the automotive world’s new V8. FSD hasn’t proven that Level5 is unachievable or that Santa doesn’t exist.

SO we all remain wishful…
 

Dids

Well-known member
First Name
Les
Joined
Dec 21, 2019
Threads
8
Messages
1,766
Reaction score
3,771
Location
Massachusetts
Vehicles
04 Tacoma, 23 Cybertruck
Occupation
Self
Country flag
“X” envy here(just sayin’)

Agreed, ADAS human-in-the-loop will always feel grounded to drivers by design; its inherent. I still lean toward KISS cruise control. Yet remain hopeful, like a kid wishing for Santa’s present, FSD will relieve my hands from the driving - physically. As Santa doesn’t exist, EM reminds - wishful thinking does.

ComSci background taught me wishful thinking is how projects go “Boeing” if they ever deliver. Which gave me to pause on FSD promise until the science catches up with its aspirations. PUREvision appears the effective pathway over radar/lidar/*dar with AGI-based awareness system.

Missing, IMHO, is additional functionality in architecture. EM loves “the thing” FSD” as a single-stack edifice. But software defined projects often start on basic premise. Then either pivot or adhoc amend the design brief to complete the project.

Andrej Karpathy might have arrived at one such point. Radar coming back is probably the tell that FSD single-stack needs an objective outside measure or perspective correction - hypervisor. That would push an AGI researcher focused on AI off the bench while olde school hands brought-in some data points for FSD scientists to “right the ship”, so to speak. PUREspeculation on my part.

All these benchmarks are battles won/lost. Its not a game of inches, stats, first advantage or bigger budget - Level5 is the sole prize. Level5 is the automotive world’s new V8. FSD hasn’t proven that Level5 is unachievable or that Santa doesn’t exist.

SO we all remain wishful…
Yes. Yes. YES Virginia he does exist.
 

rr6013

Well-known member
First Name
Rex
Joined
Apr 22, 2020
Threads
54
Messages
1,680
Reaction score
1,620
Location
Coronado Bay Panama
Website
shorttakes.substack.com
Vehicles
1997 Tahoe 2 door 4x4
Occupation
Retired software developer and heavy commercial design builder
Country flag
Yes. Yes. YES Virginia he does exist.
Elon’s definitely putting ALL Cybertruck deliveries off until Xmas. It’ll be a great media coup with EM as Santa Claus!
Sponsored

 
 




Top