ChatGPT & China: The Challenge For China When It Comes To AI
One of the topics that was derailed by the very unexpected and very dramatic "banking crisis" was a third and final installment on ChatGPT and other large language model-based products.
As you may recall, we have published two stories on this topic:
Justin, our CEO, wrote a piece talking about both the technological history of Large Language Models (LLMs) and their possible future. He argued that there will very likely be a game of leapfrog over the next few years as companies engage in an arms race for LLM supremacy.
Then we followed up by taking a look at a case study: Google and its own embarrassing struggles with their own LLM product named Bard. We tried to make a wider point that one of the least appreciated aspects of - in Justin's words - the shift from "search" to "ask" is the cost per use will rise significantly for these firms.
The implication from both pieces are that both the stakes and the cost of pursuing this technology will be very high. This will obviously curtail profits and also mean far higher levels of investment.
As users we could end up with cool new products. They might be so cool that they are even a bit scary. but they will also be far more expensive for even the largest, most wealthy tech companies to run?
As consumers this may be great but as investors it might be suboptimal. These trends will likely hurt share price performance and further damage the incredible track record of America's largest tech companies.
You can find both pieces here and here.
Today, we would like to cap our little series - at least for now - by focusing on the surprising impact that this technical leap has had in China and what it may mean for the future of AI in that country as it experiences its own unique challenges with these problems.
For as long as we can remember we have heard a lot of breathless reporting that, at some point in the future, China would likely dominate these important advances in artificial intelligence.
The reasons vary but at the heart of the argument was the fact that the close relationship between the Chinese state and Chinese tech companies would be a critical advantage in both funding and clearing the legal path for the development of some awesome new technologies.
The idea was that the Chinese artificial intelligence firms would get privileged access to capital and not need to heed the inconvenient civil liberties that hamstring Western companies. In turn, this would allow greater experimentation and greater access to data and especially private personal health data to build larger, more robust models and eventually as a direct result, better products.
We suspect you have read some of these pieces but here are some examples:
At the time it seemed like something that was far too pat. It made complete sense in a superficial sense - China is investing a lot in AI, they do have very close relations between state and private firms, they do not care about civil liberties or privacy and are determined to dominate this space - but this thesis was also way too linear. The world, like financial markets are only very, very rarely linear.
Flash forward to the ChatBot wars unleashed by OpenAI's ChatGPT and suddenly China isn't just struggling but it is confronted with the fact that all the money in the world may not be enough to compete with these new advances.
There are numerous problems, some of which are admittedly not China's - or at least not their tech companies' - fault.
A list might include:
The first is that the available corpus of high quality text in Chinese that is both online and accessible is significantly less compared to English. Furthermore, this will be unlikely to change. After all, Chinese academics and authors are frequently producing their published work in English, the world's lingua franca.
Further, Chinese tech companies are now forbidden from receiving the latest generation of semiconductor chips, such as Nvidia's A800, that are a prerequisite for doing a lot of the most advanced work.
Next, there is a real censorship problem. We in the West have our own problems with chat bots lying, hallucinating, being racist and just generally being "weird" but they pale in comparison to a country that needs to censor - already and just in general life - the likes of Winnie the Pooh, Time Travel and Christmas. That is a whole other level of, well, weird.
Last and most importantly perhaps, there is the chilling effect that the recent political repression on the tech sector could have on the ability of China's fabulous entrepreneurs to take the risks to innovate swiftly. Surveillance plus censorship is a tough environment to innovate in, especially when the consequences are severe.
You can level a lot of criticism at Western or American society, technology companies and our incredibly vicious and politicized debates about censorship - and this newsletter frequently does - but they pale in comparison to what is possible in China. Censorship is not just possible but an everyday occurrence. Namely, extra judicial repression and authoritarian re-writing of the laws and norms around technology, business and much besides.
This is a political system where people, even rich, connected and powerful people, frequently disappear and the rule of law is subject to political priorities that is difficult to even imagine as an American.
Now all of this doesn't mean that China won't have chat bots nor that they won't have daring firms that take risks and make real advances in artificial intelligence. But it does suggest that it might be quite a bit more difficult to dominate a space if you spend far more time trying to control and de-risk a product or a technology than you do on actually innovating.
Every engineering cycle spent optimizing for political prerogatives is one less spent optimizing your product for the ultimate consumer.
It is hard not to admire the Chinese tech entrepreneurs taking normal business risks with the added threat that you could subsequently either be thrown in jail or simply disappear because of an inadvertent mistake from a large and incredibly complex system.
Part of the realities of doing work at the very frontier of any field of scientific advancement is accepting that a lot of mistakes will be made. Anyone who has used or read about ChatGPT3 or 4 will be familiar with this phenomenon. These are cool products but they are not flawless and have plenty of areas still to improve.
Moreover, it is also less than clear to us that you can censor Chat bots in the same way that you can a messaging platform or a social media company or even a private group chat. We are getting rapidly away from our expertise but we may find out if the Chinese Communist Party will allow a technology that can't be controlled to flourish. On present evidence the answer under Xi Jinping would likely be: no.
But perhaps that is wrong. Perhaps these difficulties will simply force the CCP and the Chinese state to realize the futility of controlling all information in an age of "ask" rather than search?
In other words, it is early days, but could the advent of ChatGPT mean the end of the "Great Firewall?"
We would be very humble about our understanding of China's infamous system to censor the Chinese internet and keep tabs on its netizens. However, no less than the originator of the system, Fang Bingxing, has been giving interviews suggesting that, thanks to ChatGPT and other tools like it, the time of the Great Firewall had passed.
“Now it’s simply software used in an online chat-like scenario. If it’s incorporated into robots and cars, we need to stay vigilant for the potential harm they could do to humans.”
So, perhaps there will be a silver lining here. China, even China under Xi Jinping, has a track record of stunning about faces when policies no longer make any sense. You saw this belatedly and chaotically on "Zero Covid" and perhaps you could see it on the Great Firewall.
In order to compete, it may be necessary to let LLMs like ChatGPT exist in China which will make it impossible to control information as they have in the past.
Therefore the real question is, what does a China under the Chinese Communist Party look like with the internet and without the Chinese internet controls its citizens have nearly always experienced even before we get to the profound and permanent impact that LLMs may have....
Exciting times and perhaps somewhat nervous as well.
*******
Have questions? Care to find out more? Feel free to reach out at contact@pebble.finance or join our Slack community to meet more like-minded individuals and see what we are talking about today. All are welcome.
