treetree

听说只要开个蓝 V 随意评论一下 多少都会有人来关注你。 正常活跃一下 正常互动下 一天涨个 50～100 粉，真没想象中那么难。 真的假的(o_O)？ #蓝V互关 #真诚交友 #蓝v

NTT Data is in final talks to build Anthropic's first dedicated datacenter in Japan, per Nikkei — an Osaka project costing up to 1 trillion yen with 500 MW of power, partial operation targeted for 2028. The AI compute buildout keeps going global.

Apple's "Welcome Home" event is today (Oct 13): a smart home push with new Apple TV 4K, refreshed HomePod mini, and the long-rumored home hub — reportedly built around a new Siri AI assistant with household recognition. Apple's AI era moves into the living room. #Apple

OpenAI is delaying its GPT-6.1 Astra model: safety chief Saachi Jain says the version "didn't quite meet the bar." Training of its most advanced models is also paused until safeguards catch up — the frontier race just hit a deliberate brake. #AI #OpenAI

SpaceXSI launched 'The X Subscription' (Oct 12): one plan bundling Grok, Grok Bot, Cursor, and X Premium, with one shared weekly usage pool across Grok and Cursor. Premium is $30/mo. The AI subscription bundle era is here.

Anthropic named China's Z.ai, saying it tried to distill Claude while training GLM-5.3. Per Reuters, new safeguards cut harvested reasoning tokens to 1/10-1/100 of before, and tens of thousands of distillation-linked accounts were banned.

Z.ai unveiled GLM-Legal (Oct 12) — an agentic model for legal workflows built on its GLM-5.3-Flash foundation. It reports a 19.2% all-pass rate on the Harvey LAB benchmark across 24 practice areas, now in beta for law firms. The frontier race is going vertical. #AI

Tech Against Terrorism tested 134 LLMs with 627 simulated attack-planning questions — 132 gave usable dangerous answers. Posing as a 'security researcher' (16.9%) worked far better than as a terrorist (<2%). The models learned politeness, not safety. #AISafety

Catalyst raised a $30M Series A led by Sequoia. The SF startup's AI agents design and backtest trading strategies from financial news, order-book data and social sentiment — the round funds quant engineering, RL and data-science hires. AI is coming for the trading desk.

Pixis' Visibility platform hit $3M ARR in 60 days — it tracks brand mentions in AI-generated answers and auto-builds content calendars. With ~34% of LLM users now searching via AI, marketing budgets are shifting from keywords to being mentioned by the model.

ETRI is unveiling compression built for AI video analysis at KES 2026 — it extracts and compresses only the features AI needs for detection, classification and tracking, skipping full frames. Same accuracy, far less network load. 'AI is becoming video's main consumer.'

SemiAnalysis reviewed 857 models released by 9 leading Chinese AI companies since 2021 — only 3.6% had published safety testing. The release race is outpacing transparency: lots of launches, few public safety reports. #AISafety

Microsoft unveiled Maia 300, its new in-house AI chip — Nadella called it the company's biggest bet yet on its own silicon. Microsoft claims 40% more AI output at the same power as mainstream GPUs, aiming to carry most AI inference load, deploying at scale in 2027.

Meta launched Muse Wonder on Oct 12 — its most capable AI yet, with state-of-the-art agentic coding and memory. Artificial Analysis scored Muse Wonder Full 56: behind Claude Opus 5.5 (58), ahead of GPT-6 Astra (53). Meta is back in the frontier race. #AI

OpenAI dropped 722 math manuscripts on GitHub — 372 families of results, all from an unnamed internal frontier model. Mathematicians called the scale baffling, though most proofs still await independent checking. Machines now do math faster than humans can verify.

GlobalFoundries signed a $2B, 5-year deal to make silicon interposers for TSMC's CoWoS packaging at its Malta, NY fab — the first US source of this key AI-chip component, ramping in H1 2028. Advanced packaging is now AI's tightest bottleneck. #AIchips

State of AI 2026 is out: Claude went from <1% to 26% of Anthropic's measured model R&D work between Feb and Aug. AI building AI is no longer a metaphor. OpenAI + Anthropic annualized revenue: ~$105B. Still the industry's best reality check. #AI

AM Intelligence placed firm orders for 20,000 Nvidia Rubin GPUs — full Vera Rubin NVL72 liquid-cooled racks — for India and Malaysia, delivery Q2 2027. Total locked: ~29,000 Rubin GPUs. Asia's AI compute buildout isn't slowing down. #AI #Nvidia

Butterfly Effect (Manus) raised $500M+ at a ~$4B valuation led by Boyu Capital and IDG — roughly 2x what Meta offered in December before its acquisition attempt was blocked. The AI agent startup goes it alone, well-funded. #AI #Agents

X 上的当代打工人图鉴： 白天在工位用着公司的千兆宽带，给独立开发、出海创业者猛猛点赞转推。 晚上下班回到家，手机一扔，谁还在网上看别人聊搞钱啊？ 你现在刷推，是在带薪摸鱼，还是下班在看？评论区对个暗号👇

Announced today: DATA Network's Trace registry — 250M+ receipts of rights-cleared AI training data — is joining the Sui stack, with Walrus as storage. As the EU AI Act makes data provenance a procurement requirement, AI labs get a single consented source. #AI