ChinAI #354: Industry Gossip - overdue training fee payments and overhyped embodied AI
Greetings from a world where…
Overview
Greetings from a world where…
Jamie Ding from Jeopardy isn’t my cousin but I’m rooting for him like he is
…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).
Feature Translation: Leiphone Detectives Vol. 4 [AI情报局]
Context: One of my favorite things about following the NBA is the constant churn of the rumor mill. It’s been a while since we caught up on some industry gossip, so this new Leiphone Detectives series [AI情报局] caught my eye. One nice feature of these reports: each industry rumor concludes with a plug for people interested in additional stories on the topic, linked to the relevant reporter’s WeChat ID.
Key Takeaways. Rumor #1: A leading LLM company owed a cloud vendor over 100 million RMB in training fees last year.
The LLM firm owed a cloud vendor based in southern China (Jeff’s guess: Huawei or Alibaba Cloud?), but eventually paid the debt back “one kuai (yuan) at a time.”
Apparently, overdue payments are an industry open secret. Another LLM firm couldn’t pay salaries for two or three months, according to a salesperson who worked there. Still, the financial strain is lessening, as LLM start-ups expand to overseas markets.
Rumor #2: A major tech firm overhypes its embodied AI capabilities.
Previously known for its express delivery and e-commerce capabilities, a major tech firm in Beijing entered into the embodied AI track, boasting that it would collect 1 million hours of physical data in less than a year!! (Jeff’s guess: could it be Meituan? That feels wrong given the small scale of the numbers below)
According to interested clients who approached the firm, “they discovered that the company currently possesses only 30 physical robots for data collection—a number dwarfed by the clients’ own fleets.” They left deeply disappointed, as collecting 1 million hours of real-world robot data usually requires a fleet of around 1,000 physical robots.
Rumor #3: Chinese AI chip companies turn to plan B after initial failures
Over the past three years, numerous Chinese AI chip companies have actively sought out industry giants to be “incorporated into” or absorbed. Their hoped-for acquirer was a leading Chinese system integrator in the Xinchuang industry (information technology software/hardware). In some cases, negotiations got so advanced that employees from the chip company started to address their counterparts at the potential acquirer as “colleagues.”
Ultimately, all acquisition negotiations ended in failure. Now, these firms have listed on the STAR Market (Shanghai sci-tech board) or the Hong Kong Stock Exchange. (Jeff’s guess: Moore Threads and Illuvatar CoreX).
Another AI computing firm that spun off from a large tech company has had to adjust their salary compensation practices. They are also exploring a pivot to the software sector.
FULL TRANSLATION: Leiphone Detectives Vol. 4
ChinAI Links (Eight to Exchange)
I’ve been behind on my reading and recommendations, so we’ll double our dosage this week in an effort to catch up.
Must-read: China AI Bulletin
For the Safe AI Forum, Emmie Hine provides biweekly updates on AI development, governance, and safety in China. This was an extremely informative and insightful issue, including summaries of 88 AI-safety-related papers published by Chinese researchers as well as coverage of MIIT’s standardization effort for the Model Context Protocol.
Must-read: China’s Agentic AI Controversy
Samm Sacks’s Lawfare piece unpacks the uproar around ByteDance’s Doubao AI phone. She finds that the AI agent (embedded into the phone’s operating system) has “sparked a flurry of discussion among China’s legal scholars and technologists about how regulators should address the unprecedented challenges agentic AI poses to data security and privacy.” This report is chock-full of references that I will be coming back to, including many links to commentaries from Chinese scholars.
Should-read: The Trade War Lab
Based at the University of Kansas, the Trade War Lab aims to decode the US-China trade war “from the bottom up.” It’s a very cool model that brings together undergraduate and graduate students with faculty mentors to conduct research. For instance, this issue examines China’s support for liberal order enforcement (sanctions) based on UN Security Council voting data.
Should-read: My Tesla Was Driving Itself Perfectly—Until It Crashed
In The Atlantic, the former head of Uber’s self-driving car division, Raffi Krikorian tries to piece together what went wrong with his own accident while operating a Tesla in full self-driving mode.
Should-read: A Signal Point of Failure: Integrating BeiDou into U.S. Positioning, Navigation, and Timing Systems
While doing some research into the rise of devices that can receive signals from multiple global navigation satellite systems, I stumbled upon this War on the Rocks op-ed by Jesse Humpal, a former NSC staffer in the Biden administration. These days, it’s rare to come across a genuinely creative policy idea: “integrating China’s BeiDou satellite navigation system as a backup for civilian use to enhance redundancy and interoperability across the private sector, including in critical infrastructure.”
Should-read: China Rolls Out Interim Regulations on AI Human-Like Interaction Services: A Detailed Analysis
Geopolitechs provides an analysis of China’s newly-issued Interim Measures for the Management of Anthropomorphic AI Interaction Services, highlighting a narrowed regulatory scope from draft to final version (from any AI service that combines anthromorphic features to a more qualified “sustained emotional interaction services”).
Two things I did with Johns Hopkins SAIS Institute for America, China, and the Future of Global Affairs
The AI Race is Many Marathons, Not One Sprint: A memo that argues the U.S. has a healthy lead in the AI diffusion marathon, which suggests the focus should be not shooting yourself in the foot.
Panel on Tech, Rivalry, and Competing Visions of the Future: alongside Mieke Eoyang, Samm Sacks, and Selina Xu, I shared some thoughts on “The China Debate We’re Not Having.”
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.
Details
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).
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99
Source
Originally published at chinai.substack.com.
