I’m trying to stay informed about the fastest changes in China’s AI technology and policy, but I’m having trouble finding reliable recent news. If anyone has trustworthy sources or summaries on the latest developments, it would really help me understand what’s going on and why it’s important.
So in the last couple months, China’s AI scene has been moving at, like, warp speed. Baidu dropped a new version of Ernie Bot, which is kinda their ChatGPT rival. Supposedly got close to GPT-4 in some tests, at least according to Chinese press. Whether that’s hype or not—who knows. Then there’s ByteDance, who is all in on building their own LLM after catching a little FOMO from their Western competitors.
Government-wise, big stuff: In March 2024, China’s internet regulator—the CAC—tightened requirements for “generative AI” products. If companies wanna launch chatbots or whatever, they gotta do security reviews, make sure everything is “socially healthy,” and toe the party line. Self-censorship is basically baked in. Last month, they did a big seminar to talk up “safe, responsible” AI, but it’s always a lil’ vague where the red lines are until someone crosses ’em.
Export controls? Oh boy, there’s still a ton of drama around Nvidia chips. The US is squeezing export rules tighter, so now companies like Huawei are going all-in on making their own AI accelerators—think Ascend chips—and rumor is, they’re catching up surprisingly quick. Meanwhile, Chinese AI firms are snapping up every GPU they can get, even if it means buying laptops and ripping out the graphics cards (no joke).
Pro tip for sources: MIT Technology Review and The Information have pretty good China AI coverage, if you don’t mind paywalls. Also, South China Morning Post’s tech desk usually posts updates on AI trends and policy, and their stuff is less US-centric than, say, Wired or TechCrunch. For academic takes, SupChina and Sinocism do weekly roundups; they’re not real-time, but solid for bigger picture.
TL;DR: Chinese firms are racing to match (or beat) Western models, government is clamping down to make sure nobody steps out of line, and the hardware race just got a new turbo boost because of US policy. Don’t blink, things are moving fast—or, y’know, at the speed that gets the okay from Beijing.
I honestly have a bit of a different read from @reveurdenuit, especially around the “catching up to GPT-4” claims and the real-world impact of those policy shifts. Not to say China isn’t hustling—they definitely are—but every time I see state media or company hype about “Ernie is as good as GPT-4!” I kinda take it with a huge, salt-mine-sized grain of salt. When you dig into the technical benchmarks outside of China, things still seem a step behind, both on sheer model capability and on the openness/collaboration side. ByteDance and Alibaba talk a big game but, so far at least, their models just aren’t as accessible for pure testing as OpenAI’s or Google’s.
On policy, yeah the CAC keeps shifting its “red lines,” but it’s almost like the rules are designed to be flexible levers; I wouldn’t say it’s just self-censorship—it’s like a moving target of what’s allowed. China wants “controllable AI,” but that makes genuine open-source or community-driven improvement pretty much impossible—compare that to what’s happening with LLaMA or Mistral in Europe. And re: Nvidia: plenty of talk about homegrown Ascend chips, but China’s still importing as many black-market A100s as possible (according to Bloomberg’s recent coverage), so I wouldn’t count US export controls as a slam dunk for Chinese chip self-sufficiency yet.
As for trustworthy sources, @reveurdenuit is totally right about SCMP and MIT Tech Review. But I’d be very careful with South China Morning Post after their recent editorial shifts (they’ve gotten a bit more “aligned” since Alibaba took over, at least IMO). For academic/industry stuff, ChinaAI Newsletter on Substack is pretty legit—though it’s sometimes a bit too bullish on “China soon” narratives for my taste. TechNode is decent for business moves if you filter the press releases out. For actual Chinese coverage/rumors (if you can translate), 36kr and Pandaily are surprisingly useful, though, fair warning, it’s all wechat/baijiahao sourced so don’t expect western journalism standards.
TL;DR: China’s AI is moving fast on paper, but there’s a lot more smoke and mirrors than cold reality so far, and the chip race ain’t a done deal. Just don’t take company press releases or state media benchmarks at face value. If you want super granular, occasionally spicy takes, Ellen Huet at Bloomberg and Lillian Li on her Substack are both worth following. But, yeah, keeping up with China’s AI means reading between the lines—literally.