Well, @realGeorgeHotz *is* betting that Lombard on a rainy night and Bombay in heavy traffic are all fine. That’s just what the bet is, global L5, not L4, disregarding regulatory issues but pending serous safety data. You are on my side, he isn’t.
SAFETY
-
Defining True Autonomy: No Steering Wheel Required
By
–
It’s No now and no it shall remain for longer than many people think. I have been saying this since 2016… the criteria are clear: you don’t need a steering wheel or a driver and you can go where you want to go; anything less is (at most) L4, by definition
-
Phoenix Testing Inadequate for Full Level 5 Autonomous Validation
By
–
a major city is not adequate for L5, since eg you can pick Phoenix w good weather and few pedestrians which is just a subset of the full problem. See eg
-
Level 5 Autonomy Challenges in Urban Environments Like NYC
By
–
No, that is not L5. There may be some cities like that, especially ones designed from scratch, but that’s not L5. And it won’t help eg in NYC. Certainly possible in principle to be safer than humans, fully automated, and it probably will happen, I just doubt it will be soon.
-
Autonomous Vehicle Safety Evaluation and Independent Agency Oversight
By
–
Most commutes, around the globe, really? I am willing to bracket regulatory issues and pulled over vehicles if we can agree on a way to evaluate safety that is some kind of independent agency that has access to data.
-
L5 AGI Safety Betting: Can We Achieve It by 2030?
By
–
You want to put money on (safe as humans) L5 by end of decade? @metaculus @MatthewJBar https://
x.com/realGeorgeHotz
/realGeorgeHotz/status/1591567145032368129
… -
Innate Priors Shape AI Learning: Evolution Over Minimalism
By
–
– you still need some prior to organize that infinite data and decide how to generalize from it
– the choice to minimize is an aesthetic choice, not a scientific one; evolution has clearly endowed many animals with significant priors (eg baby ibex climbing down a mountain). -
Balancing AI Risks and Benefits: A Regulatory Approach
By
–
Correct. The risks outweigh the benefits. I don’t think it should be illegal but it should be approached as a drug or a loaded weapon.
-
Strong innate priors in machine learning systems
By
–
that’s learning that the set of relevant cases is empty; it’s still a strong innate prior
-
Controlled Perturbation: Function Application as Risk Factor
By
–
But in reality you may fail when you try to do f(X), so doing it is effectively just another perturbation, albeit under your control.