The situation at Cruise is so similar to OpenAI (political removal due to prioritizing success over safety) that the coincidental timing is almost contrived.
At least in this case, Vogt is doing a clear resignation on his own terms.
At least in this case, Vogt is doing a clear resignation on his own terms.
Cruise is an extremely tough spot right now after its handling of the San Francisco accident, and it's nationwide pause of testing.
OpenAI has been extremely successful but is having strategic disagreements within its non-profit board.
Anyway Altman gives me the ick, serious lizard person vibes.
> At approximately 9:30 p.m. on October 2, a human-driven vehicle struck a pedestrian while traveling in the lane immediately to the left of a Cruise AV. The initial impact was severe and launched the pedestrian directly in front of the AV. The AV then braked aggressively to minimize the impact. The driver of the other vehicle fled the scene, and at the request of the police the AV was kept in place.
Why was cruise under fire for this?
Furthermore, Cruise attempted to cover up the fact that the car dragged the woman. Now that I refresh myself on the details of the situation, it seems you have the "cleaned" PR statement that was pushed by Cruise and had omitted the above details
https://www.latimes.com/business/story/2023-10-26/cruise-rob...
Oh well. Maybe next decade.
And no, most human drivers would easily detect a human being stuck under their vehicle within seconds.
There's a good chance the woman would have been _killed outright_ if it hadn't been a Cruise. Are we willing to accept strange and inhuman failure modes if they overall lead to better safety than human drivers?
https://www.sfchronicle.com/bayarea/article/woman-run-autono...
It should be "was the eventual outcome better than what a human driver would have achieved in the same situation?"
It's not actually clear that a human driver would have done better, and given that the human driver in this incident did a hit-and-run and still hasn't been caught, assuming that is... questionable.
The safety record of these cars is actually quite good when you consider the environment and times of day they operate in.
I don't really see the parallel.
It seems impossible to have success in self-driving without safety. Cars that are unsafe will literally kill people, which will then kill the program as seen with Uber.
The AI safety/success tradeoff seems far more nuanced. The threats are abstract, largely hypothetical, and safety seems to be directly in opposition with success. So it's easy to see why product-oriented people would feel justified in making a call to ignore or down-prioritize safety, in a way that nobody running a self-driving car program can afford to.