Great and insightful read, thanks Dean! I particularly appreciated the distinction between technology innovation and diffusion. I think you’re making an important omission on the policy perspective though. Reading the article makes it seem like China has no AI or data policy constraints, while the West has many. That’s not true. China has merely set its policy objectives differently, and executes them at different points in time.
An alternative hypothesis is that the Chinese government has the power and willingness to enact policy objectives at a later stage, even when innovations and companies turn successful. So it doesn’t need to constrain ideas early. The West on the other hand struggles with this, as companies, once successful, start lobbying for power and influence to turn the environment into their favour. That means that boundaries need to be negotiated early in the West, as short-term profit maximising objectives trump all other goals and long term gains & societal benefit. Consequently, those in favour of regulation (I’m one of them), will aim to set the rules as early as possible - perhaps to the detriment of innovation and diffusion - because once the cat is out of the bag, there’s no turning back.
I think that's exactly right. My impression is that China leverages the lack of the rule of law to domestic advantage--ie, they are fine with a political economy that basically says, "you can do what you want, unless you piss us off." They also seem much more willing to tolerate capitalistic dynamism than western societies. But I don't know a ton about Chinese tech policy or political economy, so I decided I'd simply not mention it much in this piece (though I do in one sentence).
You said you haven't mentioned the European Union because of it's regulatory regime, but have you actually looked at the performance of models from companies based there or just decided to look at the regulation? In terms of model leadership, on all public arenas and by modality:
Web Agents: Global leader is Convergence (UK based) ahead of Manus and with more users than Operator
Voice: Global leader is ElevenLabs (UK based, European roots)
Image Diffusion: Global leaders are Recraft (UK), then BlackForestLabs (Germany) and Stable Diffusion is top 5 (UK)
it's a very bold argument that the EU, which doesn't house almost any of the companies you have mentioned, is ahead of anyone in terms of AI innovation or diffusion, but I encourage you to make it! perhaps I am wrong. you haven't made it here, though.
Look at open-source leader boards for any of those modalities and you will find UK/EU based teams. Perhaps local market adoption gives the impression diffusion has happened less in the EU but I can't see how you can ignore the role they've played globally ( Grok uses BlackForestLabs for imagegen, or did at one point, for instance. Many big platforms use ElevenLabs for voice agents, etc).
Yes I get that. But your suggestion that the EU policy stance has hindered innovation is provably incorrect, as evidenced by the companies above, some of which are global leaders in the (unassociated) field of diffusion models.
Things I can't *easily* do in 2025, (*with a tool that everybody just naturally knows about and is using):
* Order a chipotle burrito by typing "get me a chicken burrito with medium salsa" into a text box.
* Add a calendar entry to my google calendar by copy pasting an email text (e.g. "This thursday at 7pm we will have socer practice") into a text box.
* Extract a to-do list from my last 100 Gmail emails by typing a command or request into a text box
Where is the application layer? Where is the application layer? Where is the application layer? Where is the application layer?
It's 2025, how many more years??? Google? Anybody?
Exactly!
Great and insightful read, thanks Dean! I particularly appreciated the distinction between technology innovation and diffusion. I think you’re making an important omission on the policy perspective though. Reading the article makes it seem like China has no AI or data policy constraints, while the West has many. That’s not true. China has merely set its policy objectives differently, and executes them at different points in time.
An alternative hypothesis is that the Chinese government has the power and willingness to enact policy objectives at a later stage, even when innovations and companies turn successful. So it doesn’t need to constrain ideas early. The West on the other hand struggles with this, as companies, once successful, start lobbying for power and influence to turn the environment into their favour. That means that boundaries need to be negotiated early in the West, as short-term profit maximising objectives trump all other goals and long term gains & societal benefit. Consequently, those in favour of regulation (I’m one of them), will aim to set the rules as early as possible - perhaps to the detriment of innovation and diffusion - because once the cat is out of the bag, there’s no turning back.
I think that's exactly right. My impression is that China leverages the lack of the rule of law to domestic advantage--ie, they are fine with a political economy that basically says, "you can do what you want, unless you piss us off." They also seem much more willing to tolerate capitalistic dynamism than western societies. But I don't know a ton about Chinese tech policy or political economy, so I decided I'd simply not mention it much in this piece (though I do in one sentence).
You said you haven't mentioned the European Union because of it's regulatory regime, but have you actually looked at the performance of models from companies based there or just decided to look at the regulation? In terms of model leadership, on all public arenas and by modality:
Web Agents: Global leader is Convergence (UK based) ahead of Manus and with more users than Operator
Voice: Global leader is ElevenLabs (UK based, European roots)
Image Diffusion: Global leaders are Recraft (UK), then BlackForestLabs (Germany) and Stable Diffusion is top 5 (UK)
Many more cases.
it's a very bold argument that the EU, which doesn't house almost any of the companies you have mentioned, is ahead of anyone in terms of AI innovation or diffusion, but I encourage you to make it! perhaps I am wrong. you haven't made it here, though.
Diffusion methods were invented in EU! Any public leader board has EU or UK headquartered companies in the lead for quality and usage for image gen.
You can also add:
Tabular Foundation Models: Global leader: PriorLabs (Germany)
Avatars (Models + application to be fair): Global leader: Synthesia, (UK)
Just to be clear, I am referring to the concept of diffusion in economics and technology studies, not diffusion models.
Look at open-source leader boards for any of those modalities and you will find UK/EU based teams. Perhaps local market adoption gives the impression diffusion has happened less in the EU but I can't see how you can ignore the role they've played globally ( Grok uses BlackForestLabs for imagegen, or did at one point, for instance. Many big platforms use ElevenLabs for voice agents, etc).
I am not referring in any way to diffusion models. https://en.wikipedia.org/wiki/Diffusion_of_innovations
Yes I get that. But your suggestion that the EU policy stance has hindered innovation is provably incorrect, as evidenced by the companies above, some of which are global leaders in the (unassociated) field of diffusion models.