When you click on links to various merchants on this site and make a purchase, this can result in this site earning a commission. Affiliate programs and affiliations include, but are not limited to, the eBay Partner Network.
I take Jalopnik as serious as I take the National Enquirer. I actually want to apologize to the National Enquirer 🤣
lol I love the National Enquirer the content is so extremely stupid and over the top it's entertaining. Jalopnik is just a hate filled rage click site it's downright sad.
lol I love the National Enquirer the content is so extremely stupid and over the top it's entertaining. Jalopnik is just a hate filled rage click site it's downright sad.
That's why I apologized to the National Enquirer....they know they are full of it while Jalopnik is full of it
The Optimus thing IMO is deception by omission. Tesla didn't claim that they were fully autonomous, but they didn't make any effort to clarify that they weren't and are of course banking on observers coming to their own favorable conclusions.
exactly. playing to the believers.
Originally Posted by LeX2K
Jalopnik is just a hate filled rage click site it's downright sad.
Cognition means knowing, Elon has never said Optimus knows and is as capable as a human you've made that up to fit a narrative. Optimus is learning to do tasks like a human emphasis on learning they were not in the crowd just to entertain.
Your narrative is right now Tesla's robot is not what Elon says the ambitions are for Optimus. You expect Elon and Tesla to go from zero to perfect finished product no iterations. Next you're going to say that Elon trying to get to Mars is a failure because it hasn't happened. Maybe it will never happen but it won't be for lack of trying, it's the trying part you place no value at all in. All the hard work the teams at Elon's various company do, it seems to have zero value to you.
Fundamentally that’s not what embodied cognition mean. You keep on comparing remote controlled bots to them learning what humans do - no, they weren’t leaning anything in that so called demo and it’s not the same as FSD. Let me give you an example
Tesla FSD - Take the incident where it mistook a white truck trailer for the sky and caused an accident—that’s a clear failure in object recognition, and the fix? A software update. FSD relies on patches to correct errors, so it’s not adapting on its own. When in use - it learns nothing.
Now, compare that to embodied cognitive androids. These systems don’t just wait for updates when they make a mistake—they learn from their environment in real time. If an android drops something fragile, it adjusts for next time, getting smarter with each interaction. FSD is focused on one task, driving, while androids are built to handle a wide range of complex, real-world scenarios.
I believe that as well, but I don't think that's necessarily bad, I remember Apple rigged the first iPhone demo (you can look this up), which had the true believers in a frenzy. Now look at the iPhone. Optimus has potential, which will eventually come to fruition, although probably not on Tesla stated time lines.
I always thought Scotty should level down the meth use
Fundamentally that’s not what embodied cognition mean. You keep on comparing remote controlled bots to them learning what humans do - no, they weren’t leaning anything in that so called demo and it’s not the same as FSD.
It is the same, Optimus and FSD are part of the same training stack. As I said before you have a base level misunderstanding of how Tesla's AI works. Optimus runs a version of Tesla's AI computer, and it relies on only vision sound familiar? Tactile sensory input will be done but it's not there yet.
Let me give you an example
Tesla FSD - Take the incident where it mistook a white truck trailer for the sky and caused an accident—that’s a clear failure in object recognition, and the fix? A software update. FSD relies on patches to correct errors, so it’s not adapting on its own. When in use - it learns nothing.
Now, compare that to embodied cognitive androids. These systems don’t just wait for updates when they make a mistake—they learn from their environment in real time. If an android drops something fragile, it adjusts for next time, getting smarter with each interaction. FSD is focused on one task, driving, while androids are built to handle a wide range of complex, real-world scenarios.
You are confusing what is run locally on each node (car or bot) versus what is happening at the data center. There are no humanoid robots with even a tiny fraction of the compute and general resources on board to do what you are suggesting. One day yes.
It is odd to me that you think driving isn't a complex task and isn't a real world scenario. Androids are aspirationally able to handle any task a human can and learn as a human does, in fact that seems to be what you expect Optimus to be today or it's a fraud, a failure, a remote controlled trick. You appear to have zero tolerance for the inbetween phase, the actual process of achieving the set goals.
It is the same, Optimus and FSD are part of the same training stack.
how do you know this?
Optimus runs a version of Tesla's AI computer, and it relies on only vision sound familiar?
how do you know this, but that's a pretty obvious conclusion, use the tech you already have but the use cases are ENTIRELY different. optimus right now appears to be a really nice piece of mechanical design, borne out of elon musk's love of sci-fi. even if it's remotely controlled, it's still impressive that it can move as it does without falling over, etc. presumably it has some ability to react to balance on its own for example (gyro's, accelerometers, etc), but we don't really know that. it could be 'playing' a recording of sensors on a human doing those movements without much else (any?) smarts.
You are confusing what is run locally on each node (car or bot) versus what is happening at the data center.
i could be wrong, but i don't believe fsd upgrades dynamically, they make a new release from rules and gathered data and then release it pretty much like a phone app update. not sure the data center has any role in fsd while it's moving down the road ??
It is odd to me that you think driving isn't a complex task and isn't a real world scenario.
i know not aimed at me, but obviously driving is a hugely complicated task with big consequences for getting it wrong. tesla is making steady progress it seems which is great. they will keep learning from experience and refining it to make fewer and fewer mistakes. at some point it will be good enough (make no mistakes) for the vast majority of driving. for the other 1% the car will say "sorry charlie brown, i give up, you drive"
how do you know this, but that's a pretty obvious conclusion, use the tech you already have but the use cases are ENTIRELY different.
They are not entirely different in fact they share many similarities. Vision input, output to movement. The same inference chips and computer are in Tesla cars and Optimus. The same, and the same base software stack. Think about it, you use your eyes and brains to drive a car, pick things up, sort things, move things. You're delineating things that shouldn't be.
optimus right now appears to be a really nice piece of mechanical design, borne out of elon musk's love of sci-fi. even if it's remotely controlled, it's still impressive that it can move as it does without falling over, etc. presumably it has some ability to react to balance on its own for example (gyro's, accelerometers, etc), but we don't really know that. it could be 'playing' a recording of sensors on a human doing those movements without much else (any?) smarts.
We KNOW that Optimus is learning via vision there is an entire lab of robots doing tasks and same robots learning via humans controlling them That is how Optimus works, all of this and more is freely available information.
i could be wrong, but i don't believe fsd upgrades dynamically,
It does not, there isn't enough processing power or memory to do that.
they make a new release from rules and gathered data and then release it pretty much like a phone app update. not sure the data center has any role in fsd while it's moving down the road ??
Data collection is quasi real time but a one way street, telemetry goes to the mother ship gets compiled/trained and becomes a new version of FSD. You don't drive around on the same version and it learns and gets better.
i know not aimed at me, but obviously driving is a hugely complicated task with big consequences for getting it wrong. tesla is making steady progress it seems which is great. they will keep learning from experience and refining it to make fewer and fewer mistakes. at some point it will be good enough (make no mistakes) for the vast majority of driving. for the other 1% the car will say "sorry charlie brown, i give up, you drive"
No one actually knows when FSD will be X amount better than a human whatever factor is needed to be in wide release. Maybe it will never be, maybe in a year. No one knows. Can't compare services like Waymo because they are using explicit coding to drive it's a completely different approach. Tesla was doing this but gave up.
I think it's crystal clear the visions of these two companies is drastically different.
No one actually knows when FSD will be X amount better than a human whatever factor is needed to be in wide release. Maybe it will never be, maybe in a year. No one knows.
i truly don't believe it will be never. there will always be edge cases it can't handle. tesla and regulators will just have to decide when it's good 'enough' and how it handles circumstances where it's not sure (flashers, slow down, pull over, alert passenger/driver, etc).
Can't compare services like Waymo because they are using explicit coding to drive it's a completely different approach. Tesla was doing this but gave up.
I think it's crystal clear the visions of these two companies is drastically different.
It’s not a question of different visions, what you’re seeing above is different execution. And right now, Waymo is ahead. They have been operating fully autonomous driverless vehicles and delivering robotaxi services for quite some time. Tesla’s future edge will be in data, their customers are providing them with massive amounts of data, but in the present waymo etc have the execution edge.
If Tesla tightly confined themselves geographically like Waymo they would be acing FSD, having it work basically anywhere increases difficulty exponentially. And then there is cost, 1 Waymo is $150,000 or more plus the hardware is gigantic no consumer would ever buy a car that looked like a Waymo or would accept a pile of wires and computers in the trunk. Very very different ambitions Tesla is going for generalized autonomy on every car, anywhere.
At the end of the day it just has to work. Customers don't care much how you got there or what the prototypes looked like. Waymo is about to open up in Los Angeles and Austin in addition to SF and Phoenix. Their philosophy appears to be if it works in LA it can work anywhere, and that's not at all unreasonable. It's certainly a valid alternative to Tesla's philosophy. To be clear I want Tesla to succeed and am confident they will. My main point is that this industry is advancing fast and it's not just Tesla that's betting the future on autonomy.
