ChinAI #245: Around the Horn (13th edition)

Greetings from a world where…

ham balls and potato oles reign supreme

…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).

Around the Horn (13th edition)

It hasn’t been too long since the last Around the Horn issue (ChinAI #239), but my initial scan through WeChat surfaced so many cool articles, so let’s do another round. For ChinAI newbies, here’s how it works:

• I give short previews of ten articles related to ChinAI (all published within the past week or so). The title for each preview links to the original article in Chinese.

• Readers choose next week’s feature translation by replying to the email and/or commenting on the post with the number of their favorite article. *I give a little added weight to readers who financially support ChinAI through paid subscriptions.

• The idea here is that any of these 10 links would have made for a great feature translation this week — like Cardi B’s Invasion of Privacy, there’s no skips!

1) Landing large models, what can vector databases do?

Summary: Diffusion-oriented stories always catch my attention. This one tackles the data limitations for implementing large language models in China, specifically the role of vector databases, which store text or images alongside their vector embeddings: “data’s numerical representation as a long list of numbers that captures the original data object’s semantic meaning.”

Source: AI科技评论(aitechtalk) — focuses on in-depth reports on developments in the AI industry and academia.

2) When romance in science fiction movies comes true

Summary: In August 2023, Chinese smartphone company OPPO announced that its AI assistant, Xiaobu, would be based on OPPO’s own AI model, Andes GPT. This longform piece digs into the history behind Xiaobu, with profiles of key engineers such as OPPO’s chief software architect, Hansheng Hong.

Source: 人物 (renwu) — magazine that covers human-interest stories. A previous issue translated their investigative report of the algorithmic pressures faced by Chinese delivery drivers.

3) The protection network for “digital natives” has been woven, but it still faces implementation challenges

Summary: We should also take an adoption-oriented perspective toward China’s digital policies and AI regulations. This article takes on this task by analyzing implementation challenges in China’s efforts to regulate Internet usage by minors, especially when it comes to video game addiction.

Source: 财经E法 (Caijing ELaw) —portal focused on internet governance under the umbrella of Caijing Magazine, a respected business platform.

4) Alibaba’s DAMO Academy confronts the “King of Cancer”

Summary: It’s important to make the case for international collaboration (or, U.S.-China coupling) in AI; it’s even more important to give tangible examples of why cooperation is good. For example: making progress against pancreatic cancer, one of the most deadly and hardest to detect cancers. This article tracks the efforts of DAMO Academy’s medical AI team in this space, which produced this Nature Medicine article, with collaborators from Harvard and Johns Hopkins.

Source: 雷峰网 (Leiphone) — media portal that covers China’s science and tech landscape, with a focus on AI-related happenings.

5) GPT-4 Turbo’s Chinese benchmark evaluation is out!

Summary: Every month or so, I check on SuperCLUE, the most comprehensive Chinese-language evaluation benchmark for large models. The latest update finds that OpenAI’s GPT-4 Turbo takes the top spot, further widening the gap to more than 30 points with the leading models from Chinese labs.

Source: CLUE中文语言理解测评基准 (SuperCLUE) — organization that tests the capabilities of large language models from Chinese and international labs.

6) Hu Yong｜Regarding the risks of artificial intelligence, global regulation takes the first step!

Summary: Hu Yong, a Professor at Peking University’s journalism school and an influential blogger (who frequently criticizes the Chinese government for privacy infringements), gives his take on major developments in global AI governance, including the Bletchley Declaration from the UK AI safety summit.

Source: Digital Economy and Society (数字经济与社会) — recent ChinAI issue translated debates over China’s recent draft generative AI regulations on this platform

7) Blue Paper Report on Large Model Governance (2023)

Summary: Teaming up with the Chinese Academy of Sciences institute of computing, the China Academy of Information and Communications Technology (CAICT) released this report on China’s approach to large model governance, with a focus on the main implementation tools and suggestions for improvements.

Source: 中国信通院 (CAICT) — a think tank under China’s Ministry of Industry and Information Technology.

8) Watch out for Japan’s Technonationalism

Summary: This article analyzes a shift in Japan’s semiconductor policy from inward-looking development toward an outward-facing strategy based on international alliances, such as the TSMC-Sony factory in Kumamoto and cooperative efforts with the Netherlands and the U.S. to impose semiconductor controls on China. Some interesting takes on how Japan’s interests may diverge from those of its allies.

Source: 文化纵横 (Wenhua Zongheng) — leading platform for contemporary political and cultural thought, which also publishes a quarterly journal.

9) AI Taiyan — using intelligent tech to explore the mysteries of classical Chinese

Summary: I flagged this piece in last week’s “Four to Forward.” It’s a fun exploration of large language models’ understanding of classical Chinese, following a LLM specifically tuned for comprehending classical Chinese texts by Beijing Normal University researchers.

Source: 章黄国学 (Zhanghuang Guoxue) — public account dedicated to Chinese Studies, based out of Beijing Normal University.

10) In the first year of large models, the all-powerful Taobao has omnipotent AI

Summary: QbitAI sometimes does these in-depth, lengthy industry reports. Using evidence from China’s biggest online shopping festival (Double 11, November 11th), this report studies the use of AI in lowering the barrier to entry for merchants on Taobao, the most popular e-commerce platform (e.g., virtual assistants, and AI-generated product images and descriptions).

Source: 量子位 (QbitAI) — news portal that regularly covers AI issues, similar to Leiphone and xinzhiyuan. Lately, QbitAI has been publishing longer reports more frequently.

ChinAI Links (Four to Forward)

I’ve thrown a lot of links and text at you. Let’s close out with an abridged version of “Four to Forward,” with short blurbs.

• By the Numbers: Tracking The AI Executive Order: keeping track of the 150 requirements of the White House Executive Order on AI, a great tool launched by a team at Stanford’s Human-Centered AI institute, including past ChinAI contributor Caroline Meinhardt,

• DeepMind laughs in the face of AI policy control methods with a distributed training technique (relevant paper): Can you train large language models distributed clusters, instead of a single centralized one? H/t to Graham Webster for recommending this paper featured in Jack Clark’s ImportAI newsletter.

• Foreign direct investment is exiting China, new data show: for PIIE, Nicholas R. Lardy analyzes the latest trends in FDI. H/t to Martijn Rasser for sharing.

• A year in the life of the Ohio State-Michigan rivalry through The Blade's obituaries: Why did we lead off so many ChinAI issues with a college football tracker? Maybe a little bit of the answer can be found in this very cool approach to covering the Ohio State-Michigan rivalry, from a newspaper that covers northwest Ohio and southeast Michigan.

Thank you for reading and engaging.

These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.

Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).

Also! Listen to narrations of the ChinAI Newsletter in podcast format here.

Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99