The big picture: AI's production-era growing pains are showing all at once — a lawyer is fined $5,000 for citing AI-hallucinated witnesses, courts are starting to put a price on hallucinations, and Meta is forced to change its AI over invasive questions. Meanwhile capital keeps pouring in: a world-model startup is raising another ¥1 billion, and robot-training-data firm Mecka AI is nearing a $500M valuation.
🔥 Top 3
1. Zhejiang University grads building world models, raising another ¥1 billion
A startup founded by Zhejiang University alumni from the same lab as DeepSeek's Liang Wenfeng is closing a ¥1 billion (~$140M) round for world models. As the LLM narrative gets crowded, "AI that understands the physical world" is becoming Chinese capital's next big bet — likely the largest world-model funding in China this year. Source
2. Lawyer fined $5K over AI-hallucinated witnesses in a murder case
A New Mexico lawyer cited ChatGPT-generated testimony from witnesses who never existed in a murder appeal — and the court fined him $5,000. AI hallucination just got a price tag in criminal justice: from fabricated case law to invented witnesses, verification duties for AI content are hardening from advice into obligation. Source
3. Mecka AI nears $500M valuation in Sequoia-led deal amid rush for robot training data
With Sequoia leading the round, Mecka AI is approaching a $500M valuation as robotics companies scramble for real-world training data. Collecting data itself has become the new pick-and-shovel business — with humanoids heading to mass production, high-quality real-machine data is now scarcer than compute. Source
🌐 Around the AI World (7 stories)
Kimi-maker Moonshot AI targets $2B in annual revenue — one of China's leading LLM labs is now putting a public number on commercialization, as revenue targets and IPO rumors heat up together.
India's Pocket FM doubles revenue run rate to $500M as AI powers 93% of audio content — AI isn't just a cost tool here, it is the content engine itself; the "AI-native content platform" model has proven unit economics in emerging markets.
Meta says it's changing AI suggestions after posing invasive personal questions — after users exposed Meta AI volunteering awkward, prying questions, the company is reworking its suggestion logic; the personalization-vs-privacy debate is replaying in conversational AI.
Bank agent goes live for 42 million small-business owners — credit, bills and tax services in one place: financial institutions are moving agents from customer-service pilots into core business at scale, making finance one of the fastest verticals for agent deployment.
State Council: further improve computing infrastructure and push key technology R&D — top-level policy backing for compute continues; from "east data west computing" to a national unified compute network, policy supply keeps compounding.
Inspur plans up to ¥9B private placement to build AI infrastructure — the AI-server leader is writing its capacity expansion directly into a placement prospectus; the supply chain is betting in unison that compute demand keeps compounding.
Slack can now vibe-code interactive charts and reports inside chats — Salesforce brings "talk your reports into being" into the workflow: collaboration software is becoming the delivery surface for enterprise AI, and the BI form-factor race restarts.
📊 Trends to Watch
Distillation turns from accusation into open strategy debate: yesterday Anthropic called out Chinese labs for distilling its models (see previous issue); today Y Combinator's Garry Tan urged US open-weight labs to "distill frontier models, too." The fight over knowledge distillation is shifting from a compliance question to competitive strategy — the battle over model IP boundaries has only started. Source
AI safety enters its "real users" stress test: Claude users found ways around safeguards for bioweapons research, capping a week in which Anthropic was repeatedly in hot water over cybersecurity. Capability is shipping faster than guardrails harden — standing red-teaming is becoming the default operating mode for frontier labs. Source
Disclaimer: For reference only; exercise caution for investment or decisions.