_China_Chatbot_01
What happens when Chinese media meets Chinese AI
Hello there! Welcome to China Chatbot, a twice-monthly bulletin under Lingua Sinica that will give you a selection of all the weird, worrying and wonderful things going on right now at the intersection of Chinese media and AI.
Because there’s a lot going on. Industry and research bases have been at full tilt since Li Qiang, China’s premier, announced “AI+” at this year’s Two Sessions. The aim is to not only innovate with Artificial Intelligence, but push it into every area of Chinese society and industry.
That means bureaucrats, boffins and businessmen alike are dragging broadcasting, social media, propaganda, books, film, and all the rest, into the Fourth Industrial Revolution.
Things are moving fast, so many new policies, LLMs, and jargon coming through each day now that it’s hard to keep up with them (let alone ‘grok’ them).
But hopefully we can help with that. For each bulletin I’ll put in a selection of the hot, hilarious, interesting or damning things I’ve been seeing this month. This first one is coming to you for free, but will be for paid subscribers only going forward. Enjoy!
Alex Colville (Researcher, CMP)
_IN_OUR_FEEDS(4):Education Augmentation
At a conference on June 29, academics from both Beijing Normal and East China Normal Universities launched a guide for how students can use LLMs when writing essays. The guidelines (unavailable online) are the first of their kind in China according to The Paper, codifying AI into the learning experience. The Paper has them as saying students must say which part of their work was written by AI, and that a bot write no more than 20% of it.
TL;DR: Universities across the world are on the fence about AI-generated essays, but some in China are following the government’s lead and proactively embracing them
New Phone Who Dis?
Douyin’s “Blackboard Newspaper” (抖音黑板报), an online safety public-info account, recently posted a video warning of scammers using face-changing AI. The video showed a sweet Chinese uncle on a video call with his son, who frantically tells him he just ran someone over and needs help paying the surgery bills. But the son is a changeling, a miscreant wearing a digital mask to con the old man. The video advises users to ask suspicious-looking friends and relatives to pinch their faces to check if they are made of flesh or binary. While it’s clear the AI doppelganger in this video is a real-life actor (AI-generated humans are still noticeably stiff), Douyin is clearly taking no chances.
TL;DR: Douyin expects a wave of scammers abusing tech that will be evermore realistic and accessible
AI Weighs Anchor at the Grassroots
China’s latest local ICC unveiled its AI anchor — and it’s slick for a third-tier city. He’s the digital child of Weifang Bohai International Communication Center in Shandong and China Daily, a Beijing-based outlet under the State Council Information Office. The anchor, named Douglas, is an uploaded, polished version of real-life China Daily journalist Douglas Dueno. AI Douglas does weekly broadcasts, and says he can speak seven languages. China’s international propaganda has always struggled to get foreigners to relate to their clunky but inviolable political slogans. Weifang’s ICC hopes to take a step away from that, saying it will use the anchor to tell the city’s story well abroad.
TL;DR: Central media is injecting cutting-edge tech right down to the grassroots level, helping Party goals to tell China’s story well
OpenAI Closed to China
OpenAI has pulled the plug on China. Developers in China started receiving messages on June 25 saying that from July 9 they would lose access to OpenAI’s API, along with developers in North Korea and Russia. Although the Chinese government has blocked ChatGPT’s website, they kept access open to the API — a socket for techies to plug into OpenAI’s powerful software and turbo-charge home-grown AI businesses. Financial magazine Chinese Business Network (第一财经) ran an op-ed saying it was a “pleasant surprise” for domestic LLM companies, who are often overlooked by developers in favor of OpenAI. A now censored WeChat post from commentator Ni Ren (倪刃) said it risked China becoming an “isolated island,” cut off from tech developments. But Microsoft announced on WeChat it would keep access to the API open to Chinese developers.
TL;DR: Businesses in China and the US are grappling with how to compete with rivals and align with geopolitics. Some think you can only pick one
_EXPLAINER:CCTV Listening Media Big Model (央视听媒体大模型)
That’s the worst name I’ve ever heard.
Yep, I don’t think they were expecting it to get translated to English.
Wait, who's “they”? What is this thing?
They are CCTV — China’s state broadcaster — and it’s an LLM trained for them by a prominent AI lab down in Shanghai. CCTV provided the training data, and the Shanghai lab the algorithm, launching the LLM jointly in July last year.
OK, so what?
The model can generate scripts, voiceovers, images and AI anchors — everything you need to create an entire video from scratch. CCTV used it to create a cartoon series in February on traditional Chinese poets, generating all the imagery and movement for them. Apparently, all they had to do to create a character was upload a picture for reference and then write prompts like “Chinese style, Tang Dynasty, middle-aged, handsome, male, light green clothes” and the model would change it for them. It halved production times, they say.
Huh cool, but that’s still not much.
Yeah, I think CCTV agrees. Looks like the series was a test drive to see how the model fared. Now they’ve roped in some of the brightest and best in Chinese tech — researchers at both Tsinghua and Peking Universities, and developers from iFLYTEK and Zhipu AI — to help them develop it further, but it’s still in the R&D phase. Seems like there may have been some teething problems… When they launched the cartoon series in February they said it would be 26 episodes long, but only the first six came out.
Why are you telling me about an LLM that hasn’t done anything for four months?
Because the powers that be at CCTV are throwing some of the country’s smartest people at this, and are talking about it as a future staple in making and editing their video content. If they get it where they want it, the model could help the station produce a lot more content, faster.
_ONE_PROMPT_PROMPT:Chats with Chinese LLMs (trained as they are in Party propaganda) are both an eye onto how the CCP sees the world and a rabbit hole of weirdness.
I’ll round off each newsletter with a keeper.
(Source: Tencent’s 元宝)