> like a road with traffic, cars parked, pedestrians about, weather

Not all of those need to be done "quickly". That's where LLMs fail

You note the weather when you leave. You understand the traffic five minutes ahead. You recognize pedestrians far ahead of time.

Computers can process a lot in fractions of a second. Humans can recognize context over many minutes.

The Waymo may have done better in the fraction of a second, but humans can avoid being in that situation to begin with.

Computers can take all of those things into account as well