Tesla Admits Current ‘Full Self-Driving Beta’ Will Always Be Level 2

AI means so many things that it has become as meaningful as "magic". Some don't consider anything less than human-level an AI (this is also called AGI or strong AI). On the other hand enemy behavior in video games is commonly called "AI" even when it is as simple as pacman ghosts. In the same way that magic can literally mean "impossible" or be applied to card tricks accessible to 5 year olds.

They’re basically long-term 0% loans like the Cybertruck deposits. They keep it in escrow and slowly recognize it feature by feature.

It seems strange considering the wide availability of information and communication, but apparently there either is no feedback loop back to the people who write like that, or there is but it's mostly positive (be it more sales, less lawsuits or a greater following). At some point the meaning of those words will have hollowed out so much that we can't use them in normal conversation anymore.

Well, yes? That's how subsets work.

Thought experiment, pretend "City Streets" was called "Level 2 Mode". Would you be upset that "Level 2 Mode" will always be level 2? Or that "Level 2 Mode" is a subset of "Full Self Driving"?

I'm a robotics person though so it was always pretty clear to me that I couldn't possibly think of being able to deploy an L4 system with their set of sensors.

Prediction:

* Ridiculous Self Driving

* Ludicrous(+) Self Driving

* Plaid Self Driving

People who buy it know what they are getting, and if they don't, it's entirely willful ignorance. I don't even own a Tesla yet I've read plenty about the actual capabilities and the fact that these cars can't truly drive themselves.

Sort of like self-deprecating humor might not be the entire picture of mental health, but a sign that it is probably ok.

EDIT: I think if Elon Musk gets too serious, or gets shut down, I suspect some ingredient of entrepreneurial spirit might be lost. The "do the right thing poorly" vs "do the wrong thing correctly"

My only concern has been the stability of the connection, there are still intermittant disconnects happening on a daily basis, and it comes out to <3 minutes of downtime per day. If that's a dealbreaker, it might be best to hold off until they get it sorted out. Besides that, it's a great deal. The dish is dead simple to install, and even the router they provide is pretty damn good (sadly it only provides a single RJ45, so keep that in mind)

I still have the occasional drop out... I can't make it through a video conference, or wifi call without a freeze up or audio drop. My Starlink app says "No obstructions" but also says "Obstructed for 4 minutes" on a regular basis. Many people seem to echo this. I'm not honestly sure if this is what to expect or not.

I generally switch to my 1.5Mbps DSL line for phone calls and video conferences, which lowers the quality a ton, but at least I don't have to ask people to repeat the last 30 seconds. Web surfing and downloading large files is blazing fast, and multiple people sharing the link seems to have no impact on the quality. Video streaming is fantastic. Another downside is that you're behind a CGNAT, so you'll never get a public IP address. Then their router puts you behind another NAT...

StarLink is better than nothing for sure. But if Comcast didn't want $25,000 to run 550 feet of cable to my house, I would probably use that instead. I now have to call in an arborist or someone and top some trees that I am guessing are causing temporary obstructions, but I'm not really sure. I have a lot of trees on my property.

* Achieved dominance in the commercial launch market, arguably with an even more substantial technology lead.

* Performed the first landing and reuse of an orbital rocket booster.

* Developed a heavy-lift version of that rocket that has the largest payload to LEO out of any currently operational launch vehicle. Also with reusable boosters.

* Delivered a new human spaceflight system to NASA, the first such system since the Shuttle ~4 decades ago despite decades of promises and plans to the contrary. Starliner, the incumbents' entry for the commercial crew program, has not yet flown and is not expected to until later this year at the earliest.

* Developed the first full flow staged combustion cycle engine to leave the ground under its own power.

It seems really questionable to bet against them at this point, and minimizing their accomplishments to date is really, really silly. Not sure why you have to involve the rover in your comparison, though—it's also a monumental achievement, but it's barely comparable to the domains SpaceX is working in with their launch vehicles.

There are currently about 1.5 billion cars on the planet.

They are confined to land which only makes up 29% of the earths surface.

If evenly distributed on that 29% (~57m sq miles) of earths surface (196m sq miles) there are about 25 cars per square mile.

The orbital shell in which SpaceX is flying satellites starts 550 miles up. If you confine yourself to a 2D spherical plane, this shell already has 30% more surface area than the earth, and you can use all of it, not just the 29% that is over land. If we "packed" orbit at the same density as cars across that 2D spherical plane we could fit 6.3 Billion satellites with a separation of 1000 feet minimum between any two satellites.

Of course we aren't limited to a 2D plane in space. You can fit millions of satellites in orbit with miles of separation between any 2 without issue. The bigger problem would be signal interference.

autopilot creates danger where none needed to exist by not using the full width of the lane and failing to create space through relative small speed adjustments. The cues it gives off while driving are some of the exact things I watch for when trying to identify beginning drivers so I can give them even more space.

I certainly would not describe autopilot as 'good' on highways. I'd barely call it adequate.

The issues I have with AP have little to do with their object learning or need to gather more information (though I agree camera-only isn’t enough). The car knows where the lane boundaries are but doesn’t take advantage of the width. It knows accurately enough where other vehicles are that it could avoid staying in another driver’s blind spot indefinitely. Tesla is teaching their cars how to drive with terrible habits and not building defensive driving into their rule set.

I don't expect that any self-driving systems will start out with handling all weather conditions. Even in the long term they may handle fewer conditions than humans.

It's worth noting there are many accidents caused by people driving in snow and that AI could be hugely safer at some point.

And also having to drive manually when it is snowing is probably fine for v1.

I would worry more about bicycles, they can be extremely tricky if the rider doesn't understand how to explain their presence on the road to drivers.

Until you have something like the human brain powering the car you cannot just use "two eyes"

Computers don't (yet) have the hardware nor software of a human brain. So they better come with extra tricks, like lidar.

Whether that's a mistake depends on how long that amount of time it.

For example, 10 seconds is perfectly safe. It's fine to completely ignore the road until you hear the beep, because 10 seconds is plenty of time to look around and figure out the situation and take control.

I would argue that any system that requires constant human concentration is not actually level 3.

Level 2 is the one that's dangerous by design.

Edit: Though since the main difference between level 3 and level 4 is the ability to pull over, I wouldn't be surprised if level 3 gets quickly left behind or skipped entirely by many developers.

1. Be robust to bad visibility, not an easy task in computer vision.

2. Have access to extremely detailed maps that include things like driveways and parking lots. These maps would need to be continuously updated. The labor required to make and maintain such maps will limit how many places cars can be fully autonomous.

3. Understand the etiquette of the road. For example, a person seeing a car stop in the street near a parking spot would assume the car is about to parallel park and give it room to pull off the maneuver. An autonomous vehicle wouldn't give it so much room, because it can't reason about context in that way.

4. Negotiate ambiguous situations requiring interaction with other drivers. There are frequently situations where it's not clear who has the right of way from the rules of the road. People resolve these by gesturing with their hands and with their cars. A fully autonomous vehicle would need to understand these signals.

Tesla fans know that FSD isn't ready. They know that the $10,000 upgrade doesn't actually grant FSD, but the purchase of it indicates that they believe it will in the future. You can call it a $10,000 gamble if you wish, and you can express your belief that it will never happen, but to say "it's a $10,000 option that doesn't work" is only a half-truth, IMO.