Robotaxi is close… Here’s how I know (Major Turning Point)

We've reached a major turning point in FSD -- the disengagements are now more interesting to see than "good moments," because the good moments are expected now.
This means shifting the focus of my channel to the disengagements. This is a bullish sign for FSD, because it means we're closer than ever to full autonomy.

[@yamenarhim9336]

You did a good job this you got recommended to my main feed at the top, this is a thing a lot of people want to hear and watch, nice title bro.

[@jokedog]

FSD v12.3.4 drove from Newport RI to NYC. That's 3 hours of driving in over 200 miles. It wasn't a perfect drive though as it took the wrong off ramp once, drove into the wrong driveway when it was suppose to go into the supercharging area. Got a lot of FSD is degraded due to weather msg especially when the sun is so low it blinded the front and sometimes the side PIllar B camera. A great leap forward but still far from robotaxi ready.

[@TheRaizerx]

I think this is the right move. Disengagements are way more interesting than successful drives. And having tracked these, seeing how/when they get fixed will be interesting. Thanks for this. Subscribed!

[@myhusyatemysandwich4591]

Subscribed. I am looking forward to this new phase in your content. Wish you luck in your endeavor.

[@brian-nunez]

Yeah, I’ve also noticed that FSD can handle most scenarios. I’m so happy it doesn’t freak out at most construction zones anymore. I live in Arizona and the FSD WILL NOT get on or off the HOV lane on highways. I think it doesn’t see it as a lane because of the solid white line. Once that happens, I’ll be even more hyped 🎉

[@CarlForde]

"I'm going to put my money where my mouth is..", takes a drink of soda :-) :-) :-)

[@DerekDavis213]

1 billion FSD miles driven, and *supervised* by humans for *every* *second* . How many miles has FSD driven *alone* ?

[@adastra8218]

Nice jobe with the editing,👍more fsd in a complex aria is what we want to see🙌

[@tonyfromaus7855]

Subscribed. Like your plan. Look forward to seeing it happen. Please consider a central (and closer to screen) camera position and the car display set to show all cameras. That enables viewers to see the car surrounds and kerbs etc. Thanks for showing your experience.

[@compugasm]

Oh, your personal car can be used as a taxi? I don't think I'd want this. What happens if someone throws up, vandalizes, or has a hobo soup kitchen, while using my car?

[@danhle1032]

Curious how they say this will be solved...ppl don't have respect towards robots esp in philly

[@DeusExTesla]

Yeah it's certainly possible. But there's a camera inside the car. It'd be pretty easy to track someone who does something wrong in the car. And if they try to cover up the camera, you could make the robotaxi refuse to take them until they uncover it.

[@philliptemple9841]

I find myself doing the same. Instead of watching whole drives, I get annoyed when the description doesn't give me a list of timestamps of interventions or disengagements so I can skip to them. Phillip.

[@roxter299roxter7]

I kind of predicted this. Your are going to find that the limits get pushed further and further out. It’s going to become increasingly difficult to find edge cases that it can’t handle. Like a mattress falling off a truck on the highway.

[@benoitavril4806]

Yet it happens all the time. Driving only by learning and not understanding much (that is, having a causal model if the situation) is a recipe for failure and death. Yesterday a guy has been killed by the autopilot.

[@roxter299roxter7]

@@benoitavril4806 Where was he killed and how did he die? How do you know it was autopilot if it only happened yesterday. Call BS.

[@elck3]

What's interesting is *you* could handle many of those disengagements, and more importantly you could handle them when you hadn't seen any of those situations before -- what is it about humans that make them still so much better in these unpredictable situations?

[@zencone]

In horseshoes, "close" matters. In autonomous driving, "close" doesn't cut it.

[@DeusExTesla]

Is your point that it's not worth paying attention to until it's safer than humans?

[@zencone]

@@DeusExTesla I think Tesla's FSD V 12.3.4 is totally worth subscribing to in order to take some of the load off driving. It's not perfect, and most of the time it works, but you must pay attention. I just mean they're not ready to remove steering wheels, brake, and accelerator pedals. I can't say when that day comes, but they are getting closer with each software iteration.

[@benoitavril4806]

@@DeusExTesla Safer than humans on average won't do it. All humans don't behave the same, they don't have accidents because they are incorrect from time to time, they have accidents because they break the rules. A person commiting a faute on the road can explain what he did wrong, and then learn or lie promising it won't happen again. The car doesn't understand the rules, it doesn't stop because it doesn't want to kill, it stops because it reproduces the average behavior...

[@roger_is_red]

very good video!!!

[@dewberry3043]

Subscribed bro.

[@geoff5711]

fsd should be fine.just as long as the passengers are made liable.this is fine.

[@bombombalu]

"End of next year"

[@CarlForde]

FSDS feels like a step change from FSDB. Seems it has the basics of driving down now. From watching videos from a variety of people, it seems much of what FSDS has trouble with is mapping data. Eg. speeds are wrong or navigation is incorrect. On the other hand driving has become smoother and more capable of handling traffic issues. People who say Full FSD is years away, perhaps based on how long it has take to get to this point, need to consider how quickly 12.3.3, 12.3.4 and now 12.3.5 have been released. FSD isn't a solved problem quite yet, but Tesla is acting like they are confident they know how to get there now.

[@DeusExTesla]

Agreed 1000%. Anyone who has watched a good amount of videos, or has used FSDS themselves would come to the same conclusion. Biggest issues are mapping data, and a few remaining issues it could improve (avoiding potholes, etc). But we know they can do that with more training, which they have the data and compute for. I really think it's robotaxi ready in 9-24 months. May seem like a broad prediction horizon, but given that a lot of people seem to think FSD12 is 5-10 years away, or impossible with the current tech, shows they've not used the software. Like you said, it drives great. Most of the issues now are surface level stuff.

[@thermitebanana]

Yeah, nah. You can't add lidar in software.

[@bArda26]

dude pure vision system is ok until it's not lol. I'd be shocked if Tesla actually certifies it as level 3 ever with pure vision. It will be like oh it's like 99% fine but you gotta watch it! sort of thing in my opinion at best. I'm so excited to see a "robotaxi" with no steering and only cameras as sensors

[@DeusExTesla]

Have you tried fsd 12 in person? It'll make you a believer :)

[@filippxx]

My friend, what expectation.... the software is just bad often enough that you can't rely on it. Reddit is full of people with damaged wheels because of FSD trial, and saying it can do alomost anything is useless because nobody says I can almost drive... you need to be able to drive regardless of where you are and what whether conditions you face. So, unless Tesla figures out a way to say for this use-cases, location, whatever.... you can just sleep in your seat and we take full responsibility it is still a long way from being ready.

[@DeusExTesla]

Have you used v12 at all? It's really dang good, and I don't just mean "it stays in its lane mostly". I mean like, it's passing the turing test of driving, good... Some work to be done on it for sure, but they now have all the data and compute necessary to do that work. So I don't really understand the hopelessness in your comment--it's just a matter of time now.

[@WireShark]

Is this some hype video meant to pump the stock? If not, why tf would you mention the stock before the actual review. It's ridiculous how desperate some Tesla investors are to recover their losses.

[@DeusExTesla]

The better FSD gets, the more people will pay to use it, which means better bottom line for Tesla, which would probably cause stock movement. That's all I was saying.

[@BigB29357]

Excellent voice. But is it real?

[@danhle1032]

Best way to Put ur money where ur mouth is is to show us u invest in the stock

[@DeusExTesla]

I'm not going to show my bank account but I've literally been all in for years.

[@benoitavril4806]

How much do you bet it's not coming soon? Define soon. I bet 100k. Beware, I want an actual functional product, not an unveiling. It's supposed to be a robo taxi.

[@DeusExTesla]

I think we'll see them launch their network and start doing supervised rides within a year.

[@benoitavril4806]

@@DeusExTesla You mean the drives will be supervised by actual humans or the training of the NN will be supervised? It has been supervised ML since the inception.

[@stuartburns8657]

Would you out your family in a Tesla sat in the back, and trust FSD not to kill or main? No. Is it coming soon? No. Is it another desperate shareholder and investor confidence scam? Yes

[@DeusExTesla]

@@benoitavril4806 Either one is possible. More likely supervised by a human for sure. But if they got it in a few cities/regions unsupervised, i wouldn't be completely surprised based on how fast it's improving now.

[@DeusExTesla]

@@stuartburns8657 The answer is "almost." And when you consider that FSD 12 is a new architecture, and Tesla's no longer compute constrained, the progress will move faster than it has historically. To me, that means limited rollout of the service within a year or two. What's your estimation?

[@fatcat22able]

Cope

[@DeusExTesla]

Bet you haven't used v12 at all.