A leading roboticist punctures the hype of driverless cars, LLMs, humanoids

A few weeks ago I got into our car. Without touching any controls other than the screen, it backed out of our garage and driveway, drove me to the grocery store, found a parking spot and parked. On the way back, it did the same thing without issue.

Last night my spouse was "driving," and she got frustrated by the slowing traffic ahead, so she grabbed the wheel away from the car, and in a microsecond we were nearly sideswiped. We were both so terrified! But it only reinforces the takeaway: this car drives better than we do. No human can see as far and as well as the car's cameras can.

I'm not trying to undermine his wider point, which is fair. It's just hard to reconcile the words in the title ("driverless car hype") with my daily driverless reality.

I didn't read the article because it's paywalled and I am on a phone with no easy way to get around it.

That said, those three examples (LLMs, driverless cars, and humanoid robots) are extremely different problems.

LLMs are mostly language calculators, which are extremely useful within the scope of their capabilities and when applied to a relevant problem. They are basically "solved".

Driverless cars are also close to being technically solved when used in contexts they work well in (not snow). On pre-mapped roads, with established rules and road markings that are visible.

Humanoid robots on the other hand are far from solved and I doubt will be anytime soon. This is partially because of limited training data and partially because we underestimate how much distributed intelligence there is in our bodies. My guess is humanoid robotics that come anywhere close to human level dexterity are 10-20 years away.