_China_Chatbot_05
What happens when Chinese media meets Chinese AI
Hello, and welcome to another issue of China Chatbot, where today I smirk at a Chinese company’s attempts to roast Taiwanese President Lai Ching-te, find out where the tentacles of the Party’s most important fund for AI development stretch to, and fry an Alibaba chatbot with just four words.
Before that, a quick thought that’s been bugging me about Zero-Day (零日攻擊), a hot new TV show partly funded by the Taiwanese government, raising awareness among ROC citizens about how China could go about invading the island.
The team hosted a Q&A here in Taipei last week, airing their 20-minute trailer which makes a nod to AI disinformation campaigns. Taiwan’s President gives a rousing televised address, only for aides to rush in saying the speech has been hacked, a screen showing an AI-generated President declaring war against the PRC.
The problem is it’s so obviously a fake. The President glitches and judders during this bogus announcement, still speaking even when his mouth is closed (judge for yourself here 10 minutes 38 seconds in). But the AI toolbox has already evolved way past this: some Chinese state media have ironed out stiffness almost entirely. There are tools out there now that can convincingly clone your voice after just one and a half minutes of conversation.
Here’s something pretty damn smooth I made in five minutes with Chinese app Wondershare Virbo.
Taiwan is already a hub for PRC disinformation tactics and deepfakes, lending this hijacked presidential speech the weight of real possibility and urgency. Given how fast AI deepfakes are evolving, would Zero-Day’s depiction of them be fit for any purpose by the time it airs a whole seven months from now?
I put this to two of the show’s creators (producer Cheng Hsin-mei and director Lo Ging-zim), the pair agreeing they could not keep up with AI, and would instead be focusing more on the broader trends of a PRC-driven disinformation campaign. While that’s commendable, the show risks teaching ordinary Taiwanese to only watch out for deepfakes that look like deepfakes, allowing more realistic ones to sneak under the radar.
Compare this to Douyin’s anti-deepfake video from my first Chatbot bulletin. The Chinese tech company chose to cast a real-life actor in the role of an AI doppelganger, essentially future-proofing PRC citizens for deepfakes that are almost indistinguishable from reality. Because that’s likely the way we’re headed.
And on that cheery note, on with the show. Enjoy.
Alex Colville (Researcher, China Media Project)
_IN_OUR_FEEDS(3):How’s Business?
On August 21, a report from an influential IT analysis firm said China’s generative AI market was “not significant.” IDC (owned by IDG Capital, a Beijing-based investment group known for its sharp investments in Chinese unicorns and alleged links to the PLA) reported the size of the business-facing market was only 1.76 billion RMB, saying the low valuation was due to caution and hesitation by tech companies. It cautioned that companies with LLMs needed to do R&D on them, or would fall behind. It also noted that SenseTime, Baidu, and Zhipu AI combined had locked down half the market. Its predictions for Chinese AI’s immediate future were simply “many rounds of great change.” The South China Morning Post’s coverage also noted a dearth in demand: of all generative AI mobile apps, only Baidu’s Ernie Bot (文信一言) and ByteDance’s Doubao (豆包) had more than 10 million monthly active users.
TL;DR: The government expects AI to become key drivers of productivity, but companies have cold feet about the country’s market
Smart Watch Gives Dumb Answer
On August 21, a netizen posted a video of a smartwatch insulting the Chinese. The watch, built for children by Chinese tech giant Qihoo 360, was asked, “Are Chinese people the smartest in the world?” It responded that Chinese people appear smart because of their appearance of having large brains, “but the dumb ones I admit are the dumbest in the world.” It also pointed out that core modern technology was created in the West. Despite anger from netizens, the incident generated suspiciously little traffic on Weibo. Qihoo 360’s CEO Zhou Hongyi (周鸿祎) said the watch was an old model, so “not answering questions through AI,” instead scraping them from the internet. However, the company markets the watch on JD.com as having “AI Voice Support.”
TL;DR: It’s still impossible to make any LLM say the right thing every time, but a wrong step in China’s hair-trigger political environment will be far more costly for companies than elsewhere
Chinese Burns
An AI tool made with Western technology by a Chinese company is now becoming popular in Taiwan, worrying local experts. The app, known as “Roast Master” (吐槽大帥) scans someone’s social media details to make stinging burns about them, and has become popular on Instagram with younger Taiwanese. United Daily News (聯合新聞網) reports Roast Master’s company is Chinese, piggy-backing on Western tech like Google Chrome extensions and ChatGPT to create AI assistance tools for international users. Taipei Times interviewed a string of Taiwanese experts and lawmakers, who expressed concern the app version of Roast Master could access Taiwanese citizens’ contacts, locations, and social media accounts. However, Taiwan’s Deputy Minister of Digital Affairs said the app would only be banned if there was proof it was a security threat. The app is still currently available in Taiwan.
TL;DR: Chinese companies looking abroad amid a stagnant domestic market are going to keep finding their nationality a problem
_EXPLAINER:China Internet Investment Fund (中国互联网投资基金)
Let’s just call them CIIF.
Sure.
So a national fund for all things internet-y?
Exactly. The CIIF’s about page says it wants to put money into building “a strong cyber power.”
Who’s behind it then?
It’s run under the Cyberspace Administration of China (CAC), who are tasked with enforcing the CCP’s plans for the Chinese internet.
So they’re the Party’s investment arm for anything digital?
Yeah, making them hugely important. They’ve been given a lot to play with: 100 billion RMB ($14 billion).
But that’s nothing like what China’s been giving their chip funds though.
Yeah nowhere near as much, but CIIF’s about state endorsement and control rather than how much cash it throws around. The amount of money they invest is less important than the influence they’re able to wield with even a small stake in a company. Even 1 percent gets the fund board representation and sometimes veto powers. It also signals the leadership’s interest in a business’s strategy, potential, prominence, or quality. And knowing how to capture the interest of the powers that be is still key to the fortunes of an ambitious Chinese start-up. CIIF has put money into over 100 different tech companies, and since the crackdown on private tech companies from late 2020 it has expanded to AliBaba and digital media companies like KuaiShou, ByteDance, and branches of Sina Weibo.
Ok but what’s this got to do with AI?
Everything. Xi announced in his explanation of the Third Plenum that large enterprises should become the “backbone” of tech innovation. They are to be the dynamo that drives China forward: it’s private tech companies that have been pushing the boat out with Large Language Models, the foundation of cutting-edge AI. Having a stake in these companies allows the Party to harness this dynamo.
So who are CIIF interested in?
They’ve got skin in the game with big AI providers like SenseTime (IDC says they are the second-largest provider of AI B2B services) and small ones like Zhongke Wenge. I’ve drawn up a list of the investments in the AI section of their website. Each company looks promising, already used by a plethora of Chinese and Western companies, national government bureaus, and (in the case of Unitree) both the US Marine Corps and the PLA at the same time.
_ONE_PROMPT_PROMPT:Machines are never good with paradoxes. The impossible logic is sometimes too much for their circuits to take. ChatGPT understands the assignment…
… but it leaves Alibaba’s Qwen 2.5 spluttering like an early-90s Hugh Grant.
Excellent post. Thanks