[FREE] _China_Chatbot_03
What happens when Chinese media meets Chinese AI
Hello! Welcome to China Chatbot, where today I take advice on how to tell China’s story well abroad from a bot whose job is just that, raise my eyebrows at Beijing’s ambitious plans to turn itself into an “AI native city,” and get creeped out by the strengths and weakness of one of China’s strongest AI-video generators yet.
This issue is free for all, but smash that subscribe button if you like what you see :)
Before all that, a little riddle. On July 26, CCTV released a sensual AI tour through the bustling streets of Taiwan, overlaid with the unctuous suggestion that all of this is materially Chinese. From its street signs to restaurants serving up Chinese regional cuisines, everything in Taiwan, it insists, is a reminder of “home” — home, of course, being China. QED, come back home Taiwan!
A theory I have is that CCTV used AI for close-ups of the city’s street signs and night markets because access to Taipei for on-location filming may have been complicated or impossible, being Chinese state media. When the outlet has covered the island in the past, it’s mostly been from a distance or from acquired footage.
As of the time of writing, Taiwan’s Mainland Affairs Council had not responded to our questions on whether or not CCTV is allowed to film on the island (but to be fair I asked only a day ago). If anyone has an answer and wants to put me out of my entirely self-induced misery, feel free to reach out in the comments below, or email me at alex@chinamediaproject.org.
Whatever the case, this enchanting work of pure propaganda offers a crucial glimpse at how Chinese state media plan to use AI to overcome practical challenges — and make Taiwanese, they hope, revel in anti-Taiwanese content.
Enjoy!
Alex Colville (Researcher, China Media Project)
_IN_OUR_FEEDS(3):Artificial Interpretation
A newspaper for media specialists under the CCP’s Central Propaganda Department has laid out what the Third Plenum Decision means for state media’s use of AI. The article, penned by two academics from the Minzu University of China, held up AI as one of the “new quality productive forces” (新质生产力) that the Third Plenum Decision wants developed. They said it had become the “most important” tool for this development in the media, and called for the wholesale adoption of AI proofing, editing and content generation. Editorial departments should also allow the tech to completely change the traditional way news is produced and disseminated, with AI acting as the editor and gatekeeper of a reporter’s content.
TL;DR: China's (CCP-controlled) mainstream media are betting on AI to help them push their vision of China’s confidence, prosperity and openness. That’s no guarantee it will
Action That
Two of China’s most tech-oriented cities launched AI action plans within four days of each other. On July 26 Beijing’s municipal government launched its vision to make the capital an “AI native city” in just a year and a half. A profusion of state media covered the highly ambitious AI+ Action Plan, including releasing 100 products using LLMs, creating 1000 “industry success cases,” and promising to help multiple industries, including radio and TV, plug into AI. On July 30 Shenzhen launched its own action plan with similar (if smaller) goals, with just 40 implementation scenarios.
TL;DR: China is working on how to push AI into society. Beijing leaves other Chinese cities in the dust for AI R&D, being best-placed to lead the charge, but coordinated city rivalry will push local governments to push AI companies harder
Olympic Wins
Alibaba is helping the IOC forge its AI program at the Paris Olympics, the first since the release of ChatGPT, with IOC President Thomas Bach saying the company played a “defining role” in sorting out the IOC’s AI agenda. The Global Times framed the Games as a major showcase for cutting-edge Chinese AI, reporting that Alibaba provides AI-assisted action replays while their LLM QWen (通义千问) will be provided to commentators to assist their work. The tech giant also reports that two-thirds of live coverage will be hosted by their cloud server, before being distributed all over the world. It probably helps that they’ve been a sponsor of the Games since 2017, and now have a representative in the IOC’s AI Working Group.
TL;DR: A big win for Alibaba, but gold goes to the Chinese government and their dreams of pushing Chinese AI as an innovator on the global stage
_EXPLAINER:Kling (可灵AI）
Didn’t the New York Times write something about that this week?
Yeah, the software was finally released around the world this Wednesday. The Times pointed to it as evidence Chinese AI is closing the gap with their US counterparts.
Wait, so what is this thing?
So Kling is an AI-generation LLM from social media app Kuaishou (快手). It generates images and videos from scratch, and can turn images you’ve uploaded into videos, based on any instructions you give it.
Huh cool, how is it?
Not bad at all. Chinese users have been posting fairly convincing animations of old family photos and bizarre feats of imagination. Kuaishou is so confident with its new creation, it announced it will use Kling to make a short drama (短剧).
But I’m not yet convinced Kling is truly a world-beater — when I asked it to animate some images I provided, it had substantial problems with simple physics and body movement. In an Olympics mood, I asked it to animate a photo of swimmers diving into a pool, and the results, erm, defy gravity.
Someone else posted a video it made of a gymnast on the high bars, but her morphing and merging limbs make her look possessed.
Woah yeah ok, that’s not good.
The main advantage seems to be that, unlike OpenAI’s still-unreleased Sora, Kling is available to anyone. And for paying Chinese subscribers, they’ll give you higher-quality videos and also remove all watermarks.
So if you pay more you could make better deep fakes?
True, it can do things like putting together two different photos to make people look like they’re in the same room.
But it wouldn’t be as easy as all that. Generating anything substantial would be very tedious: each video is a maximum of ten seconds long, but takes five minutes to generate. So creating a two minute-long video would take two hours (endlessly pressing Kling’s “extend by five minutes” button, then waiting another five minutes). The software also refused to use photos of President Biden and Donald Trump I uploaded for animation, or politically-positive prompts like “give me a video of Trump and Biden hugging and laughing”.
_ONE_PROMPT_PROMPT:Zhongke Wenge (中科闻歌) is an AI company helping government bureaus and outlets like China Daily improve their messaging abroad. I asked their LLM “How can we tell China’s story well to international audiences?”
The answers were specific and concise, suggesting I emphasize the values China and the rest of the world share (like peace and cooperation), highlight China’s simultaneously traditional and modern society, choose a narrative angle foreigners can accept, and harness social media to push my message further.
We at CMP have been seeing Chinese state media use these tactics for many years now.
(Source: 中科闻歌)
Hi, i recently discovered you substack and enjoy it. Keep up the interesting work.
I live in Taiwan and did a bit of research on Taiwan cctv access. Chinese cannot set up their own cctv in Taiwan and Chinese cctv technology brands are banned. But some have still been found in places like the hsinchu science park. But for more local and less location-sensitive cctvs, some are openly accessible online, so Mainlanders with a VPN can access them.
I'd be curious to see the propaganda videos, though. can you share a link?