Not do I understand anyone trusting their Tesla this much. Even the cruise control freaks out sometimes and slams on the brakes from phantom apparitions. I don’t even trust the base autosteer feature going down the highway.
Not do I understand anyone trusting their Tesla this much. Even the cruise control freaks out sometimes and slams on the brakes from phantom apparitions. I don’t even trust the base autosteer feature going down the highway.
Those blue screens occurred in dramatically less complex and better understood software, too.
I'm not saying we don't understand how FSD works, or the underlying technology. It's not a complete black box. I believe there was even a proof recently showing that these things are essentially massive decision trees and they can be deterministically unravelled.
But this is otherwise totally negligent from where I'm standing.
This might be mitigated if the result includes a confidence percentage, ie panic if confidence below threshold, but even then I'm not sure how reliable this metric is.
When we need to audit something that is more complex than we can really comprehend, can we come to reliably accurate conclusions? If we need tools to do this for us, can we be certain they’re accurate as well?
All of it feels very Icarus-like. I’d be content if we tested this stuff out in isolation for another 20 years before throwing it on public streets.