Transitioning to Tesla Vision

Claiming that Tesla is an irresponsible company is pretty strong in the absence of evidence. Do you have evidence that I can review to see in what ways Tesla vehicles are more dangerous than other vehicles?

From what I can find Tesla vehicles are significantly safer in terms of accidents per mile, deaths per mile, and they have the highest safety rating of any vehicle.

Of course I could be misled and you have evidence that suggests that Tesla's are not only more dangerous, but that the company releases vehicles so dangerous and irresponsible that it should not be in any way trusted.

I'd like to see how you came to that conclusion.

Someone is against the future here.

If I were considering buying a Tesla I would 100% wait. Whatever you think of the company, their claims about what will happen in the future are often wildly optimistic in terms of timing. Better just to wait and see what they ship and whether/how well it works.

Knowing what I know after 2 years of Tesla ownership, I would cancel any order until they actually release the feature. I was foolish enough to buy the FSD option and it's a joke. They were claiming "by the end of the year" back then. I should have asked which year. It's still a great car, but Elon Musk is a lying bullshit artist.

Probably because they planned a date to stop manufacturing radar cars, and may not be able to easily push that date back. They may have cancelled contracts, burned down supplies, re-tooled production lines etc.

Or perhaps they just don't want to give customers hardware they won't need but will still add weight and potential maintenance. Or, cynically why pay for putting radar units into cars that won't need them.

A lot of people are speculating that the current delivery delays are due to a shortage of radar units, and that this timeline was rushed forward so that they don't have to wait anymore.

Because they already made plans to drop the hardware, and the software team missed the deadline.

Likely, they want to use the those weeks as a beta period to gather data from the fleet.

I will honestly state I don't know the terms you buy a Tesla on, but it wouldn't surprise me if holding on to the vehicles impacted cash flow somehow.

I assume that ‘short period’ here is relative to the age of the universe.

Deliveries are their most-watched top-line metric, and it's likely they can't delay deliveries anymore without significantly impacting their quarter.

My highly speculative hypothesis: Tesla doesn't actually plan to sell cars capable of full FSD, at least in the near term. Partially to keep costs low and partially to keep the gate open for operating a fleet of self driving cars that are capable of full FSD.

Waymo so far seems to be inclined towards maintaining a fleet of autonomous vehicles for rides, deliveries and trucking. I suspect it's easier to justify higher hardware costs per vehicle in that case.

Consider this more a wild speculative rambling than actual thorough evaluation, so take that grain of salt (I do work in AI/vision research):

The probability that they can safely pull this off is very low. Right now progress in vision-based ML is incredible. If you look at AK's twitter feed [1], you'll be amazed almost every day. But I work in planning, and for planning detecting / segmenting objects, even recreating 3d models is not enough, you need a full contextual model of the world. Right now, I have 0 clue how this could be done, I don't think anyone else does either. Here's an example of what I mean: you see an image, and can say 'there's a person there' - that's detection. you see an image, and can sort of imagine the person's shape from another viewpoint - that's 3d reconstruction / completion. you see an image, and can imagine the person moving, obeying social and physical laws, waiting to cross a street, look at a car and decide to wait. You know trains will move on rails, trees will stay still unless cut, etc. You're mixing physics engine, statistical inference, heuristics in your head, and this gives you a full, contextual understanding of the world.

I think to get there, one approach is to focus on prediction ability: create a model that learns to 'see', then 'predict' and train both at once (see for example World Models [2]). This prediction task alone was also enough for GPT-3 to seemingly learn really good models of the world. Then the hope is, if you can predict the future really, really well, somewhere in you must be a really perception / simulation / contextual understanding model. In public research we are so incredibly far from being there, prediction ([5], [7]) and prediction-based planning works (eg [3], [4], [6]) are still in the proof-of-concept stage - we're nowhere near predicting high-res images or 3d models of the world with any accuracy. Still, I do see some chance of a GPT-scale endeavor by a well-funded group like tesla leading to some big leap forward. But I'm not holding my breath. To make things bleaker, some think prediction is not enough, and you actually need embodiment (skin-in-the-game), to truly develop good contextual models (I used to be in this camp, GPT made me reconsider). This is even further away, adding hardware, safe exploration and co. is just not something we're in any way close to. Another common thought is that there's a limit to current NN architectures, and that a drastic step is needed to get to the next level, though no one knows exactly what (moving away from feed-forward architectures and backprop, towards something like spike-time-dependent plasticity? higher level meta network coordinators, that decide when to connect trained models to each other's inputs and outputs, when to start and stop modifying their weights, etc, continually?)

You could argue that it might be possible to skip this entirely, and just focus on training an algorithm to be good at driving from start to finish, but there I'd say that we are even further from having the approaches to reach that goal. I'll leave that topic for someone else.

And if you say you don't need this, then you better have some magical over-engineered planning method that can perfectly account for all possible errors / missing info in your perception pipeline. As far as I'm aware, Tesla has no edge there. I'd expect that's the approach other big manufacturers were throwing millions at for all these years, before ML suddenly became more than just edge detection / segmentation.

[1]: https://twitter.com/ak92501?ref_src=twsrc%5Egoogle%7Ctwcamp%... [2]: https://worldmodels.github.io/ [3]: https://www.youtube.com/watch?v=w6TLbv_54GY [4]: https://arxiv.org/pdf/2012.04406.pdf [5]: https://arxiv.org/pdf/1808.06601.pdf [6]: https://arxiv.org/pdf/1812.00568.pdf [7]: https://www.youtube.com/watch?v=3UZzu4UQLcI

Note: I do think they could pull FSD on motorways in a realistic amount of time, though, because there's a lot less variation there. You could even add a little bit of external control, like road-side accident detection cameras, full walls to avoid animals / objects going in and out, etc.

From what I read, Tesla uses these to collect ground truth information for training their camera-based system.

I was talking to a lawyer friend about this the other day. I was curious where the lines really are in terms of false advertising and misleading consumers. I know, for example, that part of the case against the tobacco companies was that they misled consumers about the risks of things like "light" cigarettes.

Anyway, he just shrugged his shoulders and said, "Elon Musk."

I really hope we move back away from having a set of people who aren't subject to the same rules as everyone else. I realize that life has always been like that to some extent, but it seems worse these days than in the (recent) past.

Well, it is what the feature has been called for ages. So maybe by now it isn’t Full-Self Driving but the feature we call “Full-Self Driving”.

Tesla calls this feature "Full Self-Driving Capability" and pretends that the word "capability" is understood to mean it could eventually be capable of it, somehow.

It's like if Apple were selling a smartphone with 80 Megapixel Photo Capability, and the fine print reads:

"Devices equipped with 80MP Photo Capability do not have an 80MP sensor. The capability refers to the hardware's ability to potentially upscale images to this resolution. We don't yet ship software that enables the 80MP resolution."

(I was stupid enough to pay for this bullshit FSD feature on a Model 3. Nice car otherwise, but with Musk it's always caveat emptor.)

A self-driving car is defined as "a vehicle that is capable of sensing its environment and moving safely with little or no human input" https://en.wikipedia.org/wiki/Self-driving_car, but Tesla's Full Self-Driving feature actually requires human input. This is confusing.

"Full Speed Self Driving", as opposed to "High Speed Self Driving"

wait, or is it the other way around?

ref: https://en.wikipedia.org/wiki/USB#USB_2.0

You can see the details at https://www.tesla.com/models/design#overview:

>Basic Autopilot ...Included with every new Tesla, Autopilot enables your car to steer, accelerate and brake automatically for other vehicles and pedestrians within its lane. There to assist with the most burdensome parts of driving, Autopilot works alongside features like emergency braking, collision warning and blind-spot monitoring.

> Full Self-Driving Capability $10,000...The currently enabled features require active driver supervision and do not make the vehicle autonomous. The activation and use of these features are dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates.

The general assumption is that current FSD package is 'Beta', and being worked on. Will it ever get out of Beta? 20% chance I think. But Here's Hoping.

"undriving"

>What is Tesla going to call their eventual level 4 autonomy feature? "Actually Full Self Driving"?

Technically that would be Actually Full Self Driving Sometimes.

Certainly it is misleading, but technically correct if you check those tenses:

"these will be the first Tesla vehicles to rely on camera vision and neural net processing to deliver Autopilot, Full-Self Driving and certain active safety features."

Future tenses all the way. Someone might make them clarify this, but the clarification won't be as high profile as this release.

Do we have to relitigate this every time? Do you think we’re going to advance the conversation around this by you raising this? As someone who even only tangentially follows Tesla & no stake, it’s really tiring.

You know there are still 20yo cars driving in Europe