Checking in on Waymo

The first Waymo trip I took was on September 16, 2023 at 3:05pm. It cost $18 flat. It was right after they got into private beta - so there was no driver supervisor but things were still a bit rough around the edges. I had seen prototypes for years driving around the sleepy streets of Palo Alto and the slightly busier streets of San Francisco. I had also seen some more DIY versions when I was at Stanford. Now I was using one as a utilitarian: my goal was the destination across the city, not in the technology I was using to get there.

There was something magical in the performance: the steering wheel torquing itself, James Bond's second favourite car brand, the subtle ambient music playing in the background when you first get inside. It felt like something that shouldn't exist in the same timeline as the potholes outside. Aside from that my first experience was pretty typical. It took five minutes and I was back to looking out the window.

The self-driving landscape was far different in 2023. Cruise and Waymo were both racing to get California permits. Tesla had been shipping Hardware 4 since January and announced Hardware 5 later that year. Apple was still making progress on their own self driving car.

Now - dare I say - Waymo is the undisputed champion in the states. They're expanding to new markets and all the while expanding the amount of safety miles they have logged, which helps their case with regulators.¹

In AI and self driving, San Francisco² is so far ahead of the rest of the world it's hard to wrap my head around. I've been surprised by the amount of otherwise well-informed people I know in New York who only have passing recognition when I mention self-driving cars or the latest GPT model drop. They have the same association when I mention something on the cover of Popular Science. Aware of it but it feels more like press buzz that won't really reach the mainstream.

There are two ways to read this:

1. SF is a city of early adopters. We like what's new for newness sake, even when it's not necessarily helpful to us.
2. You have to see it to believe it.³

I'm inclined to think it's the latter. Especially with AI, it's really only sheer exposure that can teach you how others are using it and how best to use it yourself. Without that it's easy to both use it wrong and not understand how much of an accelerant it stands to be.

During that Waymo drive in 2023, it was pretty rough around the edges. Braking was choppy and it had an overly strong preference to yield to other drivers. In crowded traffic it seemed unwilling to merge unless there were multiple feet of clearance. Taking Cruises at the same time revealed something similar: jerky braking, taking turns too fast, etc. Any Uber driver was indisputably better.

Fast-forward to two years later and I don't know a single person that prefers to take Ubers anymore⁴. People might still call a Lyft but it's because of cost not because they prefer the experience. I'm sure if you look at the safety statistics Waymos will boast a healthy lead. But it's not even that aspect of driving where they excel: it's the ease of the throttle, the radius of the turns, the consistent speed. Combine that with more spacious interiors and most people are sold. Even my luddite parents are fully convinced.

The thing I keep coming back to is how fast this is all happening. In under two years cars went from what felt like some CV models and a detailed control system to driving better than most people. Reasoning models in LLMs are getting good, really good, at problem solving in closed system problems. And we haven't exhausted how far we can push transformers yet.

If Waymos are any indication, it's pretty damn far.