I initially believed the term “unsupervised” was self-explanatory, but it’s now clear that traders interpreted it in different ways. Although clarifications were added later, expectations had already diverged. At that point, resolving YES or NO would have unfairly favoured one reading over another.

To maintain fairness and integrity, N/A was the only responsible outcome. I appreciate the thoughtful engagement, but I won’t be entering further debate on this.

@GordanKnott I created a version of the question with only the unsupervised no human driver, and paid service.

https://manifold.markets/brianwang/will-tesla-launch-unsupervised-full-NttZ99ICpL?r=YnJpYW53YW5n

Registering my disagreement: Although there's some chance of Tesla's launch falling into a gray area (such as level 3 autonomy), it's much more likely to be either an unambiguous NO (they miss the deadline) or an unambiguous YES (they launch something technologically similar to Waymo's offering).

More fundamentally, I think resolving N/A because of traders complaining about ambiguities can harm fairness and integrity. In this case, no reasonable person assumed that unsupervised FSD meant level 5 autonomy. Unsupervised naturally means that you don't have to have eyes on the road ready to disengage the AI. Waymo clears that easily. And if there were ambiguity about this, traders should ask. It's possible Tesla kind of cheats here and a judgment call is needed, but (a) that's not likely, and (b) that doesn't preclude finding a fair resolution.

To be more specific, if someone were to see this market and bet it down to <1% without asking "does Waymo's offering count as unsupervised FSD?", that wouldn't be remotely reasonable. If you do that and Tesla pulls this off and surpasses Waymo and this market resolves YES, you can't cry foul because by your idiosyncratic definition of "unsupervised" it shouldn't count. That would effectively just be weaseling out of paying up on a bad bet.

(But less hypothetically, people did ask, and the market creator clarified faultlessly.)

And again, most likely of all is Tesla doesn't launch in time and none of the possible ambiguities even matter.

Another similar market:

Understandable, different people clearly have very different things in mind when they say the word "unsupervised." For me, if a human often needs to take control, that's supervision

@TiredCliche No, Waymo is the natural benchmark for markets like these and there's no real ambiguity. Again, see FAQ8 of my other market about this.

@dreev I don't think there's ambiguity- Waymo are not full self driving, level 5. They often need a human to step in. There's that video recently of the guy who got trapped in one doing circles around a parking lot over and over.

@TiredCliche We just need to ask the market creator for clarification on whether Waymo would count as unsupervised full self-driving as a paid service. I would argue that the natural answer is yes based on similar markets and what people commonly mean and expect. You would argue the opposite based on, I guess an ultraliteral meaning of "unsupervised". But either way, it was incumbent on us as traders to get that clarification before trading and the market creator should feel free to clarify according to their own judgment.

Not that this is likely to matter. If Tesla simply misses the deadline, that's an easy NO and if they leapfrog Waymo that's a pretty easy YES, presumably.

@dreev Yeah. And to me, the common, ordinary meaning of unsupervised means that a human doesn't have to come in and correct it every couple miles. Like, if I had to help out my taxi driver every couple miles, that doesn't sound like he's driving the car by himself.

So I would say that the natural answer would be no. Certainly it surprised me when I learned that Waymos had humans intervening every few miles. I assume that's why they kept that hidden, because it's embarrassing.

@TiredCliche You are confused about Waymo. Please read the FAQ on my other market.

@dreev I did read your FAQ, and I already understood that a human isn't remotely operating the gas, steering wheel, and brakes.

That's why I used the analogy of a taxi car driver that gets confused and has to ask the passenger for help every couple miles.

At no point is the passenger touching the steering wheel, nor taking manual control of the car in any way.

But if I were being driven by said confused taxi driver, I would not say he was capable of fully driving the taxi, by himself. Since I needed to help him out all the time.

@TiredCliche Thanks, sounds like we're gradually getting on the same page (no human with VR headset, for starters). Waymos are also not consulting humans while actively driving down the road. The phone-a-human feature is for when the car has to stop and doesn't know what to do because of some unusual situation like firetrucks in the road. Level 5 autonomy is the AGI of self-driving, where the car can handle all situations and all driving conditions and make all judgment calls as well as a human.

@dreev I hear you say that, but I have seen videos of Waymos contacting the human supervisors while actively driving.

@TiredCliche Are you game to create a market about that?

@dreev Less game, since companies already hide the extent that they need to use human intervention, and Elon Musk is, well, an evil man who lies a lot about his companies.

@GordanKnott Looks like you just resolved this N/A? Was that intentional? I think this is a great market and your answers to questions in the comments were all 💯 so I don't think we should N/A this. If you want to rid yourself of having to adjudicate it, we can probably get someone to take over. (Also I think it's unlikely to fall into a gray area. Almost all the probability mass is on unambiguously missing the deadline (NO) or launching something similar to Waymo (YES).)

Another similar market with a bit more leeway on timing and whichI think should be trading strictly higher: https://manifold.markets/dreev/will-tesla-count-as-a-waymo-competi

It is impossible to legally drive without a human operator, currently. Computers can't do it.

NO if:

The service requires human supervision/safety drivers

Do I assume that if there is a safety driver that isn't actually required but is present on all/any June tests then this still resolves no?

Human supervision is more difficult. I can well imagine that for June tests (if that happens) Tesla has an employee not as a safety driver but to observe perhaps mainly of whether pick up and drop off points cause any issues like traffic snarl ups. That might be done from inside cars or from following cars. Either way, is this human supervision? Also maybe it is human supervision but is not actually required for the service even though it is present on tests in June? So does this situation resolve yes or no?

@GordanKnott Above questions?

@ChristopherRandles if human supervisors are present then it's assumed they're required

Related market:

https://manifold.markets/JamesGrugett/will-tesla-serve-more-fully-autonom

Clarifying questions:

1. What if it's a fixed short route that is technically public roads but involves, say, beacons or human monitors checking for pedestrians or other ways in which, in spirit, it's essentially a testing environment?

2. What if it's like bus routes, where you can't choose an arbitrary destination, even within a fixed geographical area?

3. Can we identify a threshold below which the geographical area that the trial is limited to is too small to count as more than a testing environment?

(I'm asking these because I'm worried about the scenario where Elon Musk twists the definition of public launch as far as necessary to be able to say that it happened. Like how he maybe/sorta got Grok 3 to technically, by some benchmarks, with lots of fudging, count as the "most powerful AI in the world" for 3-10 days before Claude Sonnet 3.7 and GPT-4.5 came out. I guess Grok 3 is still tied for first on the Chatbot Arena leaderboard -- I might be being harsh here. We'll see. That's why we have these markets!)

@dreev Good comment. I am sure Tesla will not have a real L4 car by the end of 2026 but also still expecting to lose many of my anti-FSD bets due to whatever gimmick Tesla comes up with to deceive the public/investors into thinking it works better than it does.

I'm now considering making a version of this market that can resolve in between YES and NO. If Tesla doesn't launch anything this summer at all, that's a NO. If Tesla launches late, that's a partial YES. To get a full YES, they'd have to launch something on par with Waymo by June 30. Then it's just a matter of operationalizing the question of what fraction of a Waymo-scale launch Tesla's launch counts as.