China sets a hard 2027 deadline for L3 and L4 autonomous driving

China’s Ministry of Industry and Information Technology has put its mandatory national standard for L3 and L4 autonomous driving out for public comment. The draft, Safety Requirements for Intelligent and Connected Vehicle Automated Driving Systems, would replace the old voluntary standard and for the first time set a legally binding floor for L3 and L4. … Read more

Xiaomi open-sources U0, a unified model for generating robot training data

As the first unified model in the embodied-generation space, U0 has already validated generation quality, data usefulness and engineering efficiency. In the WorldArena benchmark built by Tsinghua, Peking and others, U0 scored first in the world under an anonymous code, and policies trained on its expanded data lifted task completion by 26.3 per cent in … Read more

China’s tech giants are taking the AI war to the office

Why now, and why the office The trigger looks like a Morgan Stanley note from 30 July: AI investment has entered ‘halftime’, and the market is shifting from chasing upstream compute chips to two new fronts, AI applications and supporting resources. The real reason is simpler. Tencent, long mocked as behind in AI, is now … Read more

OpenAI and Kimi are fighting the second half of the AI coding war

The barrier is falling Until 2024, training a usable coding base model cost tens of millions of dollars in corpus cleaning and pretraining. Today, with Meta’s Llama 3, Kimi and DeepSeek open models, general code ability has been commoditised. A startup no longer reinvents the wheel: take a strong open base, feed cleaned code and … Read more

Ant Group’s IJCAI papers chase AI that adapts, not just performs

Why static models fail Ant Ant’s business pushes ‘not static’ to an extreme: more than 1bn users, over 8,000 services, and payment peaks versus 3am troughs that can differ more than a hundredfold. Risk models must answer shifts in black-market tactics within hours, and global deployment has to fit each country’s different financial infrastructure. In … Read more

Fei-Fei Li and Yilun Du just co-signed a paper that skips the robot brain

The missing translation interface The paper’s core point is that today’s video models understand physics but cannot issue commands. A model fed the start of a kettle tilting will continue the scene plausibly, water pours into the cup, yet it has never done anything, and faced with a robotic arm told to move a cup … Read more

Nvidia’s easy ride is over, and the bulls have split three ways

The old rule is breaking For two years Nvidia became one of the most watched tech names on a simple premise: as long as Microsoft, Google, Amazon and Meta kept buying chips, servers and equipment, the upstream supplier kept benefiting. The market watched how much these buyers pledged to spend, because more spend meant more … Read more

Another Baidu veteran bolts for physical AI

From platform to physical world Yu joined Zhongguancun Kejin as president in September 2024 and is credited with productising its large-model platform and capturing the enterprise opportunity of ‘platform plus application plus service’, improving the company’s product capability, business scale and visibility. His departure marks another senior operator stepping from internet platforms and scaled AI … Read more

China issues its first mandatory safety standard for L3 and L4 autonomous driving

China has approved its first mandatory national safety standard for L3 and L4 autonomous driving, setting a unified bar for vehicles that can drive themselves. The Ministry of Industry and Information Technology’s standard, Safety Requirements for Autonomous Driving Systems of Intelligent and Connected Vehicles, was formally approved on 30 July and takes full effect on … Read more

DeepSeek’s ‘kill line’: why a model 85 times cheaper is resetting the default

DeepSeek V4 Flash is redrawing how models are judged. On OpenCode, a single AI coding tool, the model burned 8 trillion tokens in one day, more than the entire OpenRouter platform’s daily average of about 6.6 trillion across 400-plus models. The reason is price. DeepSeek charges about 2 yuan per million output tokens; Anthropic’s Claude … Read more