@alex_avoigt You need to have evidence when making claims.
Tesla is still a rounding error, while Waymo has a remarkable safety record over 200 million miles.
@deliprao I would think such an assertion is unjustified. What conclusion funding agencies draw is not relevant here. The hope is that they will be smart enough to understand that formal verification is only a part of what mathematicians do.
@GeniusPothead@pmddomingos Sadly neural nets alone are not too interpretable. I mean computational yeah, interpretable not. LLM actually helped here. The weights of tokens may not be interpretable, but at least the tokens produce something readable.
@promiscfx The only thing that is left is some kind of manipulation based on learned rules. So yeah, neural nets gut it all, because the logical rules and the symbols were too rough to be useful.
@pmddomingos Exactly. Symbols or not symbols was never the argument (despite what Marcus would say).
The question is always how fine the level of abstraction should be, and how to properly manipulate them.
Then need grounding in either case.
@_ianwatts_ We shall see. I so much doubt these are the final numbers. Tesla is wildly optimistic.
Then, Waymo can do two seater too. They focused on the hard part. Safety first. Cost later. Tesla does opposite. We will see.
@_ianwatts_@zachtratar Electricity usage per mile is never the metric anyway. That is dirt cheap.
It is cost of the engineering, server farms, support.
And above all it is safety. Tesla is sorely lacking evidence here.
@zachtratar@_ianwatts_ The current numbers per mile are promotional. Tesla only got a show now, no real deployment that can be measured over months, no federal permit either.