A couple of weeks ago, I joined AI researcher turned co-founder and CEO of autonomous driving startup Wayve, Alex Kendall, for a test drive on a really difficult road in London. Wayve has been on my mind for a long time, and we wrote about it in EV#443 in the context of their most recent achievement — Alex’s team has created a 9-billion parameter world model called GAIA-1 to train their autonomous driving systems. Wayve uses synthetic training data, AI-generated driving situations, and prompting to enhance the driving capabilities.
I was blown away by the experience in the passenger seat. We drove on some of the toughest roads in London, and the system navigated the traffic — humans, cars, roadworks and even a police intervention — safely, smoothly, and with no human input.
After the drive, we settled at Wayve HQ for an hour-long discussion. Paying members of Exponential View will receive our full podcast conversation tomorrow.
p(doom) for AVs is 0.
Accelerate AVs with two guard rails: no ICE and no interface to nuclear codes.
Prediction: driving in rural NorCal and rural Bali will be harder than London, but with a world model foundation, it can be far safer than now.
The interview format — random talk during ride — works well.
Good to see that Tesla FSD12 may have some competition. Scale will matter and I’m not sure if smaller entrants can challenge Tesla in this space. T has invested massively, has a very large fleet and tight vertical integration.