[ex-aws-dude]

I don't get how this solves the problem of edge cases with self driving

Even if you can generate simulated training data, don't you still have the problem where you don't even know what the edge cases you need to simulate are in the first place?

[sixo]

Well it certainly helps,doesn't it? This system is going to encounter more edge cases than a single human ever would. Hopefully the lessons from known unknowns generqlise to unknowns. And once they've been seen once they took can become part of the corpus.

[ex-aws-dude]

Right but does this just rely on some human "brainstorming" a bunch of edge cases?

It just strikes me as neverending edge case wack-a-mole

A human doesn't need to see tons of examples of tornados and elephants to know to stop the car

Doesn't that indicate some fundamental difference between the model and a human driver?

[dyauspitr]

I don’t think coming up with novel situations is all that hard. LLMs already do it in text form all the time.

[ex-aws-dude]

How big is the space of possible things a car can encounter?

It’s practically infinite, your domain is 3D space and time

You can’t just generate every possible scenario

That would be a combinatorially insane amount of data