Tesla's AI Day Event Did A Great Job Convincing Me They're Wasting Everybody's Time

[Jhodgesatmb]

The entire presentation was targeting ML experts and not you tubers and vloggers (let alone other media, auto industry or financial markets). To them they showed the history and progression of how they have attacked an immensely complex problem and how their research has moved the bar forward, and where they need to go. It was intended to whet the appetites of the best and brightest, not ignoramouses. I am certain that with that audience they hit the mark. The only reason I could see for them to livestream was to catch a few other smart ML experts, and all the commentary from others can safely be ignored.

 

[HaulingAss]

I have not ordered FSD and don’t plan to unless circumstances change. I do find putting my hands on my lap or touching the wheel periodically and watching the road while the car is driving as ridiculous as he makes it sound like but saying the effort is a waste of time because it is too difficult or because this stage of the process is not the same as final stage is short sighted. I am sure the author is the beneficiary of thousands of automations without noticing it. Many of which I am sure were quite difficult while being developed and have failed many times. My understanding is Level 5 does not need a human in the car. That is just a matter of regulations. Sending your car to pick up grandma, or take you home when you are drunk have value. One thing he has failed to mention is that self driving is not just for Transporting people. A lot of drivers are on the road to transport goods and services. A car can deliver a Robot technician to a job site. The robot need similar image processing capability to walk to the washing machine and fix it. Even Bezos will pay Musk so his Rivian vans can drive themselves. Many people may choose not to own a car because FSD taxies are much cheaper to use. Same kinda tech would be essential if you are sending ton of Robots to a new planet were they have to move around in unknown environment with other robots. There will be countless applications we can not think of. Stopping progress to a place we know we will have to be because it is difficult is what I do but the reason I can afford to chicken out is that there are people out there like Musk that won’t. It would be idiotic for me to try to stop them.

People without the experience of using Tesla's Autopilot on real-world highways for long enough to become proficient with it's use almost always fail to understand just how beneficial Autopilot really is.

They can't imagine how it would reduce fatigue on long trips (although it does so rather dramatically). They can't imagine how it increases your safety by putting more "eyes" on the road and giving you lightening fast reflexes if the driver in front slams on the brakes when you least expect it (although it does increase safety rather dramatically). And they can't imagine how it turns stop and go traffic nightmare into a relaxing pleasure-cruise and music listening session without a care in the world.

My wife and I have had two cars with FSD/Autopilot for three years and were initially surprised at just how useful it is. Now we know better than to be surprised when we arrive after a long journey feeling refreshed.

 

[BillyGee]

Maybe, but by reading the comment section he has quite a few that agree.

Most auto forums out there have a serious unfounded hatred for Tesla. They will straight up make things up if they need something to complain about. Everything will be spun in the opposite direction and they will continue to talk about how the company will go bankrupt any day now.

Even my business partner, who is a smart and forward thinking guy most of the time, thanks that the mach-e and electric F-150 are going to do Tesla in. Doesn't matter that that are less than 10% of the sales numbers and facing the same shortages without any internal production, they should just give up and sell while they're ahead.

I wonder if we should all collectively buy stock in crow harvesting companies so that when these people need to eat it we can profit off of that and pay off our cybertrucks.

 

[Ogre]

Using this logic. Why even have cruise control?

Your foot is right there, you still have to brake, accelerate to pass, and steer the car. Cruise control is a waste also.

All driver assist technology is there to take the mental load off the driver. Yes you still have to watch the road, but you don’t need to constantly monitor speed. You don’t wander out of your lane when take a sip of coffee. You don’t creep dangerously close to the car in front of you.

Munro put it perfect. “Now the driver get to look out the windows and enjoy the view “. You still have to pay attention. But there are plenty of between moments where you know the cars is fine and can relax a little.

At stoplights I don’t watch the light if there is a car in front of me anymore. I just rely on autopilot. In stop and go traffic, I might look out the windows or even a passenger I’m conversing with.

Being present and alert is not the same as having to deal with every mundane detail of driving.

 

[Deleted member 12457]

The entire presentation was targeting ML experts and not you tubers and vloggers (let alone other media, auto industry or financial markets). To them they showed the history and progression of how they have attacked an immensely complex problem and how their research has moved the bar forward, and where they need to go. It was intended to whet the appetites of the best and brightest, not ignoramouses. I am certain that with that audience they hit the mark. The only reason I could see for them to livestream was to catch a few other smart ML experts, and all the commentary from others can safely be ignored.

SF Bay Area. Used to work at LLNL and after watching a shortened video about the DOJO, especially the comparison to the TOP500 supercomputers, I got excited. To be able to use this for ML, something I wasn't involved in--btw: I'm retired but haven't lost my mind--is fascinating. I was marginally involved in one of LLNL's supercomputers, which led the TOP500 for at least a few years, so I know how many resources it takes to build and install one of these monsters. From the looks of a single(?) DOJO (never heard this term used before so I'm comparing it to a supercomputing shelf/unit, which fits together with lots of other ones) running at PFlop speeds being about a foot square, I could see how little space it could take compared to the IBM supercomputer I dealt with that was at least 50-ft long and 6-8 feet tall. I was criticized in another forum for suggesting TSMC install a small silicon manufacturing unit in the Shanghai Giga factory so Tesla could manufacture their DOJO without outside help, which I didn't say but meant limited stealing of trade secrets. Tesla has the funds to do this so why not? Anyway, even though I didn't understand a lot of the presentation, it shows the Elon and company have their act together and others have not even started.

 

[firsttruck]

The entire presentation was targeting ML experts and not you tubers and vloggers (let alone other media, auto industry or financial markets). To them they showed the history and progression of how they have attacked an immensely complex problem and how their research has moved the bar forward, and where they need to go. It was intended to whet the appetites of the best and brightest, not ignoramouses. I am certain that with that audience they hit the mark. The only reason I could see for them to livestream was to catch a few other smart ML experts, and all the commentary from others can safely be ignored.

Yup

Tesla live-streamed AI day because Elon believes in getting the best talent no matter where they currently live and regardless of their formal degrees they have. Elon has said this several times.

A Fremont in-person event limits Tesla to only those few that are privileged to live in California or U.S..

 

[TruckElectric]

The entire presentation was targeting ML experts

Starting at 1:53:10 A questioned is asked if Tesla is using ML within it's manufacturing design or other manufacturing processes. Elon: I discourage the use of ML because its really difficult, unless you have to use ML don't do it....99.9% of time you do not need it.

 

[Newton]

no comment

 

[anionic1]

They right all this as they sit typing on a highly complex electronic computer that started out a wall of rotating devices and capacitors that simply solved math problems. I am sure many stood back and said, “What a foolish endeavor. Can’t people already solve these math problems. What do we need this for?” We’ll likely everything in the authors and our life now benefits from that foolish device. The only waste of time is see here is this thread written by a very short minded individual. AI is likely going to be the most important invention in human history as it harnesses the elements and physics to aid us in our endeavors while accomplishing things our minds would never be able to.

 

[Deleted member 12457]

Starting at 1:53:10 A questioned is asked if Tesla is using ML within it's manufacturing design or other manufacturing processes. Elon: I discourage the use of ML because its really difficult, unless you have to use ML don't do it....99.9% of time you do not need it.

I agree with Elon about not using ML for manufacturing but I believe the DOJO supercomputer will only(?) be used to develop FSD, which in simple terms is a computerized version of a good driver's brain. This car brain will control the vehicle more consistently than a human and will make calculated adjustments to everything in the vehicle better than we can do--much better than a lot of people can do! What you just said is correct but saying it was a waste of time simply means you don't probably understand the significance of what was presented. I'd say 99.9999% of people in the world understand half of what was said. I understood maybe 5%, and that's pushing it, but I do understand how it will help Tesla finalize true FSD and that's all that matters. Tesla is capable of pulling together experts and free thinkers to come up with an amazing computer and software, which is lightyears ahead of anything else currently on the market.

disclaimer: Yes, I own several shares of TSLA stock but that's not why I have these opinions. Even though I'm pushing 70, ouch! I just realized that, my brain still works and realizes how far we've come since my first computer programming class in college using punch cards.

 

[Diehard]

Using this logic. Why even have cruise control?

Your foot is right there, you still have to brake, accelerate to pass, and steer the car. Cruise control is a waste also.

All driver assist technology is there to take the mental load off the driver. Yes you still have to watch the road, but you don’t need to constantly monitor speed. You don’t wander out of your lane when take a sip of coffee. You don’t creep dangerously close to the car in front of you.

Munro put it perfect. “Now the driver get to look out the windows and enjoy the view “. You still have to pay attention. But there are plenty of between moments where you know the cars is fine and can relax a little.

At stoplights I don’t watch the light if there is a car in front of me anymore. I just rely on autopilot. In stop and go traffic, I might look out the windows or even a passenger I’m conversing with.

Being present and alert is not the same as having to deal with every mundane detail of driving.

Cruise control keeps my foot from cramping in a long drive. It has a physical benefit.

I have never owned a Tesla but I do believe people are not as special as they think they are. In other words if I did own a Tesla, I would likely drive it like you and @HaulingAss drive yours and would come to the same conclusion that you have come to and appreciate the benefits of it as much as you do.

However as an outsider at this point, I still can’t help thinking there is a reason you are told you have to pay attention to the road. As long as you don’t know what exactly that reason is, any moment you are not paying attention is a crapshoot. Even Munro on the same trip that he was enjoying the view mentioned that his Tesla scared the crap out of him because it tried to take an exit that was not there (or something like that). What I am saying is that as a hypothetical experiment, if everyone did not pay attention at all we would find out that out of every 10,000,000 hours of driving one minute of it could result in death. Obviously I pulled this number out of my …. but you get my drift. The way you drive your Tesla, there is a good chance statistically speaking, nothing will happen to you but if everyone drives their Tesla the way you do, at some point that minute is going to hit someone. So the convenience you are getting is from the added risk that may seem insignificant. The author is saying by flipping the system to be a guardian angel he rather trade the 10,000,000 hours of convenience with the life lost during that one minute of system failure. For most of the folks that don’t experience the death thing, it is a hard trade to make or even believe there is a trade. But what if someone told you it is not 10,000,000 hours and it is 10,000 hours. What if it was 10 hours. That is the thing. We don’t know and those that do only tell you to pay attention. You get many of those benefits by not listening. He is saying the system is pointless if you did really listen.

 

[Ogre]

Munro was talking about FSD when he mentioned getting scared by the car taking an exit that wasn’t there. Autopilot doesn’t take off ramps. FSD is in perpetual Beta for a reason.

 

[Crissa]

And that's why Tesla is so focused on vision, Ogre. So it doesn't do stuff like that.

I just... The author's argument that driving was so simple that it shouldn't be automated was so bass-ackwards. Driving is so simple, yet it kills tens of thousands of Americans every year. Nearly all from driver error.

And then the comments vacillate between excoriating them for going through level two to saying that we shouldn't have level 5 ever. But you literally cannot prove if you have level 5 or 4 or 3 without doing level 2 first. And only Tesla is trying to do it with a fleet of a million cars.

That part where they could ask the self-driving computers on hundreds of thousands of vehicles at once to bring back scenes that looked alike? That's why Tesla will work and others have stalled. Because their data set is just hundreds of thousands of times larger.

And that's how you get the confidence to go to level 5. Per mile driven, accidents and confusions are super-rare. And yet, the computer needs to experience them to know what they look like.

It's no different than a new driver watching dash-cam video to know what not to do. And level 2 is no different than having your teen drive you around - they're gonna make mistakes. That's the point of training! They can't never drive and then suddenly be a perfect drivers.

-Crissa

 

[Ogre]

When I’m driving solo, there is a hard maximum of 2 eyes on the road and no minimum.

When Autopilot is driving with only me in the car there is a maximum of 10 eyes on the road and a minimum of 8.

Maybe those extra 8 eyes aren’t as good as an alert driver’s 2, but they are 1000% better than the zero eyes an inattentive driver has. All of us have been less than perfectly alert while driving at some time or other.

 

[Crissa]

Even if we're perfectly alert, we cannot look in all directions in once or perfectly count the number of seconds between us and the next car.

-Crissa