Lugh

What is more important on the path to AGI - scaling and more training data, or fundamental breakthroughs in AI software developed by humans?

Many people believe it is the former, though Deepseek itself arose from the latter.

If training data is to be significant then Tencent may have given itself a huge boost. WeChat is the biggest and most used app on the planet. The Google suite of products is the only thing comparable in western countries. However, Western countries are starting to split into AI walled gardens, so Google won't be able to take full advantage of all its users.

It may not matter. Some people believe scaling alone won't be enough to get to AGI. Some smart humans somewhere will need to figure out breakthroughs in AI software - and that could happen anywhere.

Although tariffs might slow things down, the ultimate destiny of the world's robotaxis is probably to be cheap, electric and made in China. This week, BYD the maker of the $9,500 Seagull hatchback said it will make Level 2 self-driving standard on all its cars, including it.

When cars this cheap are self-driving and taxis, it will mean there is little point for many people to own a car. Why, if the few hundred kms/miles you drive a month costs a fraction of car ownership?

Ryan Johnson, the developer of Culdesac, thinks this trend is already helping it, and will ripple out to change the way more and more people live in cities.

Current state of Waymo in Phoenix

• Now regularly seeing my social circle, male and female, looking to it first

• Parents now comfortable sending their kids to school and elsewhere. This is a major vibe shift. Early on, women solo riders were the loudest champions. But parents are overtaking that. Effusive praise e.g. “I have my freedom back!”

• Biggest impediment to growth is that they go slower. Which of course is because they don’t speed and don’t run red lights

• Perception that Waymo makes other drivers drive safer

• Now regularly seeing Waymo convoys

• First anecdote effect dissipating. When someone sees their first minor error from Waymo, it is jarring. But then a long time elapses until they see their second. And that builds intuition that it is rare, and points the finger at how much more common errors are from human drivers

• People are asking when they can order Waymo via either Lyft or Uber

• People seeing how fast the AI tools are improving is bringing the “Waymo right now is the worst it will ever be” conclusion

Phoenix is Waymo’s most mature market, now 8 years into public availability. It’s a big reason why we chose Phoenix (Tempe) for the first Culdesac.

The May 2023 launch of the Jaguar platform was a seminal moment in the history of AV Ridehail going mainstream. And AV Ridehail is going to drive the largest change to cities in decades.

For some time people have spoken of the concept of sovereign AI. Sovereign AI refers to a government's or organization's control over AI technologies and associated data. At the start of 2025 such an idea isn't just talk any more. It's rapidly happening.

It's most obvious in Europe. Just as the US gears up to become more autocratic, the EU has passed laws to ban the AI that enables it. This week the bloc banned AI it deems 'unacceptable risk'. Among other things, it bans AI that manipulates and deceives, targets minorities, allows biometric profiling, or predictive policing. Almost everything on the list is something American Big Tech is doing with the encouragement of the current administration. To make the point clearer, the EU is building its own AI for European governments, institutions and civil service to use.

China is building AI the equal of any, and in the case of DeepSeek, perhaps the best there is. Not only that, they are Open-Sourcing it. There's no reason to think they will slow down. In fact, China may accelerate in AI; they have a huge trove of public data to use for training that the Chinese government has recently decided to make available for the first time. China is many countries in South America and Africa's main trade and technology partner. Where that is the case they may be its main AI source too.

American Big Tech has historically been used to dominating globally, but there are all the signs that it isn't going to happen with AI.

Amazon's plans

Figure's plans

Their plans are separate, but what is significant is that they are just two companies, and the raw numbers can be so huge.

Amazon expects to soon save $10 billion a year replacing humans with robots. Amazon currently employs 1.1 million in the US. If we take the average cost of each as $50K - that's 200,000 jobs. Figure is talking about 100,000 robots.

For now, this issue is still relatively politically muted. But for how much longer?