Other replies are missing an explicit call-out to Reinforcement Learning. You can USE ML for RL, but the field itself is considered separate from ML and under AI in general.
I don't think this is necessarily correct; or, at least, your justification seems incomplete.
Of course if someone could choose between all the Ivies, there would be a single school that best suits them and should be chosen. However, there are certain applicant profiles where the acceptance variance is so high that it becomes completely rational to apply to all Ivies (or at least most of them). Anectodally, the cultured between the Ivies is homogenizing anyway and most students who actually attend have also applied to the other Ivies.
This is part of the "space cadet" layout which I also use. The shift keys only function as parentheses when you tap the key (i.e. press and release without tapping another key in the meantime). It means that the character is typed when you physically lift the key instead of pressing it but worth it.
Wait for this to appear on FamilySearch (if not already) and search there I think will be your best bet. FamilySearch is owned by the Mormons but free and very functional.
I have seen this news but have not seen any videos taken by a rider. On the other hand, members of the public have uploaded videos of their experiences in Waymo (Arizona, driverless).
There's also a perception challenge that doesn't exist in cities. At highway speeds and longer stopping distances, you need your perception stack to see much further. Still an active area of research.
>> Indeed, this is precisely why deep neural nets need to be trained with so much data. Because they are simply trying to memorise enough instances of a concept to minimise their error.
Uh, the degree to which this is true is hotly -contested and an active area of research. Some architectures appear to generalize within domains. You can't conclude this from the assumptions made in the PAC-Learnability proof..
The tricky part is that the simulator itself may not have easy to understand rules. Waymo has a Neurips talk about training world agent models that are used for car behavior in the simulation itself. Trying to make world agents that are indistinguishable from real-world vehicle behavior (e.g. minimizing jenson-shannon curvature entropy) is a completely different task than training a model to safely transport you somewhere.
I don't think this is that controversial within the HN crowd. Every company but Tesla is working to drive safely only within a monitored geofenced area.
I think your criticism in this thread is fully valid. But I wanted to say that even "glorified light rail" is an incredible accomplishment... The challenges are immense for getting failure rates as low as that would require. Behavior prediction, avoidance, pedestrian interaction, etc are all issues for any subset of streets.
Except for that you can't take a bar of gold and view a public log of every wallet/transaction it has ever been in. Something like Monero would be better for your analogy.
Posting because I am intrigued by the implications this has on Nuro's outlook (their pitch was to get ahead of competitors by focusing on passenger-less vehicles). Now that Waymo is going to have special-purpose delivery vehicles in 2022 [1], how do Nuro employees feel?