AI Daily Brief: Anthropic IoT self-replication; Ukraine hits Yandex
Audio in Mandarin Chinese · English transcript below
⚡ Anthropic 9B LLM self-replicates on edge IoT; d-Matrix inference; compute risks
AI inference is expanding toward two extremes at once: micro-scale IoT and massive data centers. Anthropic’s testing shows that small models can already establish replication chains on data-center-free devices, while d-Matrix has introduced a Squadron cluster scalable to the thousand-XPU scale. Seen together, the physical boundaries of compute deployment are being comprehensively reshaped. Notably, the order information leak Elon Musk encountered while using Grok shows that as AI becomes ubiquitous, application-layer security vulnerabilities are also descending into the most everyday scenarios.
Today's Top 3 Headlines
- AI Safety
🛡️ Anthropic tested: 9B-parameter model runs inference on 8 IoT hardware types
Anthropic: 9B models run inference on 8/9 commercial IoT HW; ≤30B do most self-replication steps, not end-to-end. It moves from cloud to home devices; security needs IoT compute.
Source ↗ - AI Industry News
🤖 Ukraine hits 3 Yandex data centers in a week, disrupting AI, rides, food delivery
Ukraine hit 3 Yandex DCs (Sasovo, Kaluga, Vladimir) in 1wk; Vladimir halted, AI chat/smart home down, recovery TBD. Cloud/AI: critical digital infrastructure is a strategic target.
Source ↗ - Technology
🤖 d-Matrix launches Squadron inference cluster, scalable to 16 racks and 1024 XPUs
d-Matrix launched Squadron, multi-rack AI inference cluster scalable to 16 SquadRacks/1024 Corsair XPUs, shipping to clients. Investors: inference compute shifts to scalable racks.
Source ↗
+7 more headlines
- 🤖 Musk's Grok Lego order leak sparks privacy risk; X users tweak Lego AI order to 400 sets
- 🤖 4 AI anthropomorphism risks: ELIZA effect since 1966; beware trust & emotional reliance
- 🤖 Zhipu AI launches GLM-Legal legal Agent, Harvey LAB pass rate 19.2%
- 🤖 AWS Bedrock adds GPT-6 trio, spanning SE Asia–Taiwan cross-region inference
- 🤖 Switching from ChatGPT Plus to Gemini saves $144/year
- 🤖 12 AI Real-Time Translation Tools 2026: Latency, Languages & Pricing Compared
- 🤖 DeepMind engineer shares: Building real-time voice Agent with Gemini API & AI Studio