Hacker News Reader

Comment by nolok | Hacker News Reader

I assume they mean take emergency corrective action when an issue arise like google with Waymo, and the general thing will still be sorta kinda self driving like tesla does (which is not like waymo, but still better than nothing) ? Because being a network latency event away from a car crash doesn't inspire me much confidence, and I don't have to sit in one of those to be affected, just to be on the same road.

View on HN

> assume they mean take emergency corrective action when an issue arise like google with Waymo, and the general thing will still be sorta kinda self driving like tesla does

Having taken Waymos and put my parents' Tesla in FSD, I wouldn't assume as much. Tesla's FSD makes mistakes more frequently, and they're bad ones, in part due to its willingness to aggressively accelerate when it thinks the road is clear.

I would describe the difference as: Waymo drives like a limo driver, carefully and smoothly, and tries to disappear for the passengers in the back seat. Tesla FSD drives like a car owner might, and tries to make the person in the front seat feel like the behaviour of their car is the same as they would do it.

I rode in a Waymo for the first time last week. The highest praise I can give it is that it's boring. It drives like a careful human with zero surprises.

Yeah it’s hilarious (frightening) when Tesla cultists say that the surprise in FSD’s behavior makes it exciting.

That attitude is why people who share the roads are tired of Tesla’s and their loyalists’ constant excuses for this stuff.

Humm I wonder if there's some truth to that, some kind of pattern matching, like "at this stretch of the road usual behaviour was: speed X mph, turning at this and this points, etc"

The more important difference is that Waymo's FSD system is SAE level 4. The Tesla FSD system that is in their cars that people can currently buy is level 2.

The difference is that the former only needs human help in rare circumstances. The latter needs human help in common circumstances.

If you base you robot taxi on a level 4 system you just need humans on standby to respond when the car knows it needs help and asks for it. If you base it on a level 2 system you need humans monitoring during routine trips to be ready to intervene when the car is about to do something stupid and has no idea that it is out of its league.

And FSD is improving everyday. The only question is when will it be good enough to run a system similar to Waymo where a remote operate an help if it gets stuck, and then when will it be good enough to not need that and be full autonomous.

Place your bet.

2018!

The people who placed that bet are something like 500x their money, so I’m sure they’re happy!

Huh?

"Fuck the customers, shareholders are happy!" So odd.

Probably never unless they get off their high horses about lidar.

Network latency is no issue when operators control cars in LLM style:

„Slowly drive around on the left of the obstacle on the right, accept crossing the opposing traffic barrier line if you are sure you will be back on your side after you drove around the obstacle.“

What happens in the event that the LLM starts halucinating and driving at 100mph in the oncoming lane?

Reply on news.ycombinator.com