Riding the Wayve with Bill Gates

Recently, I had the great privilege of taking Bill Gates out for a demo ride in a Wayve vehicle through central London. It was an incredible experience to spend time with someone who’s been at the forefront of the biggest technological transformations, including the PC revolution, the Internet revolution and now the AI revolution.

Bill was eager to get a firsthand look at what Wayve was doing. He was curious to learn how we are building an end-to-end neural network that uses self-supervised learning and reinforcement learning to build artificial intelligence that can drive, not based on a set of rules or a 3D map, but from interpreting the environment around it and making the safest decisions possible. He immediately asked many sharp questions about our training dataset, compute requirements, and how we capture the diversity of all the edge cases that we encounter in the real world, many of which he experienced in his 1-hour ride through London at rush hour.

On our ride, we discussed many of the opportunities that the age of AI will bring.

Like me, he believes that the algorithms that drive AI’s learning will continue to improve and that entire industries will reorient themselves around its use. I was thrilled to show him a glimpse of the next chapter in AI: the age of autonomy, where we will see the emergence of embodied AI and the first physical interactions between humans and AI-driven machines.

Deep learning has transformed many AI problems, most recently with large language models like OpenAI’s GPT-4. As we experience the seismic shift that these models are having on how people access information, I anticipate a similar shift with autonomous machines.

To get these systems on the road, we’re investing in data and our compute infrastructure. We’ve partnered with three of the largest delivery fleets in the UK—Ocado Group, Asda and DPD—to collect real world data and experience on their routes. We’ve also built a simulation suite called the Wayve Infinity Simulator that gives us a vast array of synthetic data to train from. We are partnering with Microsoft to implement supercomputing technologies to support the development of billion parameter foundation models for autonomous driving. As these applications advance, we are designing powerful new infrastructure capable of handling datasets consisting of images and videos on top of compute-intensive workloads for training and validation.