🌍 Thanks for following! 感谢关注
Looking forward to making friends worldwide. 期待和世界各地的朋友交朋友。
Tech 科技 | Stocks 股票 | AI | Fitness 健身 | Games 游戏 | Business 商业
Qwen Office stocked rivals Nov 16: 10 models, including Qwen4-Max, Kimi, GLM, DeepSeek, MiniMax and Step. Doubao Work opened rivals Nov 12.
They tried in August: GLM and DeepSeek up on the 14th, pulled on the 17th, no public explanation. I read this as a bid for the office tab.
@Reuters The checkable bit is a Nov 2024 Shanghai Jiao Tong paper: about 5,000 answers from OpenAI's o1, turned into 10,750 bilingual examples, then used to further train Qwen2.5-72B. A year-old data shortcut. Is "the main reason the gap closed" a finding, or one US official's read?
Hong Kong already ran Anthropic's IPO test. MiniMax-W closed Monday at HK$124.90, -18%, listing low after March's HK$1,330; cap over HK$410B to ~HK$44B. https://t.co/s8AovC0m06 fell 13.42% to HK$353.60 vs June 22's HK$2,980. Oversubscription prices the book, not later trading. My view, not advice.
@Reuters Monday's close was $156.14, down 4.12%. Beating the $108B revenue guide tonight won't clear the air. I'd rather hear whether Kress still defends the old 2027-28 spend path.
Chinese posts say Hy4 Flash "crushes GPT-6 Luna." Tencent points it at daily office work, and full Hy4 at long coding jobs. Official list is $0.070/$0.281 per 1M tokens, 224B/12B active. WorkBuddy free through Nov 30, TokenHub 50% off. The 1/20 cost cut vs Hy4 is their WorkIndex.
DeepSeek is rationing the free app and web, citing insufficient compute. Idle chats over 7 days won't continue; one input is capped near 40,000 characters, plus a daily quota. No official comment. I wouldn't call this a sudden chip cutoff. They opened billed tool APIs last week.
Chinese hook: no assets needed, let the model walk your repo and make the launch video. I read the tutorial. Codex pulls real UI, Opus 5.5 directs, Seedance 2.5 only fills blank-screen plates, Remotion puts the actual UI back in. The "one morning" post sits on a v51 pipeline.
Chinese posts call MiniMax H3-Image an open-source win. It leads Artificial Analysis's open-weights T2I board at Elo 1066; 25th on the full board.
The Nov 10 license excludes the US, EU, UK, and South Korea. That title does not license those markets to run the weights.
@Reuters Friday's close was $162.85, down 12.75% on the week. Has the board actually opened the door, or is this allies lobbying ahead of Kress's first earnings call Tuesday?
OpenAI dares to spend your money, Google only reads the books. ChatGPT Wallet is US Pro, Dots only, Face ID each buy. It wants pre-approved spend later. Gemini is read-only via Plaid. Google may train on summaries; humans may review chats. I'd let it read my bills first.
@GavinSBaker The 2.8x–8.8x–14.1x gap is "per Grok and Claude." I'd want the inventory method before calling this a clear success.
WSJ says those SEA data centers were also a cover for alleged chip smuggling. Does the 8.8x count only legal cloud access, or chips that may have left the yard?
@elonmusk@MKBHD The order page still requires $299/month to operate, on top of $29,990. That's about $17,600 over five years after the included first month. You say it should be ~$115. Will the December Austin deliveries actually bill $115, or do first owners start at $299?
@Reuters The 87% is a September-October contact survey, taken before this week's Nvidia shock. The harder number in the same report: advanced-computing debt issuance was already above $680B by October, about 31% of nonfinancial IG corporate issuance, up from under 6% in 2023.
@PalmerLuckey@sporadica Reuters (Nov 5) put about 600,000 H200s in Hong Kong, with large-scale mainland import approval still withheld. If Beijing is throttling even the chips Washington already licensed, how much of Nvidia's China upside is actually collectible?