Editorial - SiliconNoon (Page 115)

The Stack

Ant's Falcon TST 2.0 tops GIFT-Eval as global banks adopt the model

Ant International's Falcon TST 2.0 claims the top spot on GIFT-Eval, the most-watched benchmark for time-series foundation models, and it got there by chasing a harder prize than leaderboard points: forecasts that hold up in live financial markets. The Singapore-based payments arm of Ant Group says version 2.0 of its Time-Series Transformer cut mean absolute scaled error to 0.666 across the benchmark's 28 datasets, ahead of models from several global tech giants, and that forecast accuracy has h

The Guardrails

China's AI ethics reviews reach city level — and a space AI model just trained itself in orbit

Two stories today about what happens after the model is built: Beijing is wiring ethical review into how Chinese cities actually deploy AI, while one of its aerospace institutes has proved a neural network can keep learning on a satellite, 500 kilometers up. China's industry ministry says it is now organizing artificial intelligence ethics review and service pilots across multiple cities, the most concrete sign yet that the country's April rulebook for AI governance is moving from paper to prac

The Guardrails

China's industry ministry lays out its AI application playbook for the next five years

Beijing is moving from AI slogans to a work plan: the industry ministry used its first "15th Five-Year" briefing to name exactly which AI scenarios get state backing, while one of China's best-known small-model startups told staff that the next battle isn't in the cloud at all. China will spend the 2026–2030 plan period cultivating "AI+", "robotics+" and other flagship application scenarios as a matter of industrial policy. At a State Council Information Office briefing on August 26, officials

The Arena

Tencent's AI chief tells staff outlasting rivals beats moving first

Two stories today about playing the long game: Tencent's cloud boss has written an unusually candid internal essay conceding his firm looks slow on AI — while in India, a voice-AI startup just convinced Peak XV that a phone call is only where the work begins. Dowson Tong, who runs Tencent's Cloud and Smart Industries Group, has published an internal essay confronting the charge that Tencent is falling behind in AI — and arguing the race is a marathon, not a sprint. The piece appeared Tuesday ev

The Arena

ByteDance ships Doubao Work, an AI agent wired into Feishu

A busy Tuesday across Chinese tech: ByteDance gave its chatbot a real job title, Alibaba's founder put his own money behind the AI pivot he's asking investors to fund, and Tesla quietly scrubbed China from its self-driving map. ByteDance has launched Doubao Work, a standalone AI agent product built to carry entire workloads rather than answer questions. The company positions it as a new product and brand aimed squarely at productivity: given a goal, it decomposes the task, calls tools on its ow

The Frontier

Qwen3.8-27B ships a fully NVFP4 build that keeps its benchmark scores at a third of the size

A quantization team called QUASAR-QAT has released a fully 4-bit build of Qwen3.8-27B that keeps nearly all of the original model's accuracy — and it did it by retraining the quantization in, not by rounding weights after the fact. A fully NVFP4-quantized checkpoint of Qwen3.8-27B is out, and it posts scores within a rounding error of the unquantized original while cutting the download from roughly 56 GB to under 20 GB. The release comes from a group publishing under the name QUASAR-QAT, and th

The Everyday

Amazon pulls the plug on Mechanical Turk, AI's original human engine

Two quiet endpoints in the AI supply chain surfaced overnight: the platform that once supplied the human intelligence behind machine learning is switching off, and one of China's breakout models is being retired by its own maker. Amazon is shutting down Mechanical Turk on September 30, closing the 21-year-old marketplace that once supplied the human labor behind early machine learning. The company announced the closure in a notice posted to the MTurk site Tuesday, saying it regularly evaluates

The Stack

Nvidia's Vera Rubin posts a 30x agentic jump in first hardware tests

Nvidia just put real numbers on its next flagship. At Hot Chips 2026 the company published the first silicon-level benchmarks for the Vera Rubin NVL72 rack, and it ran them on DeepSeek-V4-Pro — a 1.6-trillion-parameter open-weights model — doing agentic coding work rather than canned chatbot queries. Against the current GB300 NVL72, Nvidia claims up to 30 times more throughput per megawatt and per-token costs cut by as much as 35 times. The comparison baseline matters as much as the headline nu

The Arena

Alice raises $140 million to break AI models before attackers do

Two big rounds landed tonight — one for the people who attack AI models for a living, and one for the trucks that drive themselves to the store. Alice, the AI trust-and-safety company until this year known as ActiveFence, has raised $140 million led by Apax Digital Funds — and the numbers suggest guarding models has become a real business rather than a compliance line item. Most frontier labs already pay for its adversarial testing ahead of release, according to the company, and annual recurrin

The Guardrails

China's payment industry adopts a 'Know Your Agent' rulebook for AI that spends money

While the US debate over agentic commerce stays theoretical, the world's largest mobile-payments market just wrote actual rules for it. China's Payment and Clearing Association has issued the industry's first self-regulation convention for AI-agent payments — headlined by a "Know Your Agent" (KYA) framework that extends know-your-customer checks to the software doing the spending. The convention, released August 24 under People's Bank of China guidance, binds association members handling accoun

The Guardrails

X Corp.'s cease-and-desist letters take Nitter and XCancel offline

Two of the best-known ways to read X posts without an account went dark within hours of each other — both citing letters from X Corp. The no-login era of the timeline is formally ending. Nitter, the open-source front end that let anyone browse X posts without an account, went offline after its maintainer received a cease-and-desist letter from X Corp. A notice posted on nitter.net by the project's developer, the pseudonymous zedeus, says letters sent on August 24 demand a permanent takedown of

The Everyday

Leaning on a chatbot makes you worse at spotting fake news

MIT's Media Lab ran a four-week test of whether chatbots actually make people better at telling real news from fake. The answer cuts both ways — and the downside lingers after the chatbot is gone. People who used an AI assistant to check news got better at spotting fakes while they used it — and measurably worse once they stopped. In the study, 67 participants judged whether pairs of headlines and images were real over four weeks: with chatbot help their accuracy on suspect items rose 21%, but

See all