📰 𝗜𝗖𝗬𝗠𝗜: 𝗧𝘂𝗲, 𝗢𝗰𝘁 𝟮𝟳 𝗔𝗜 𝗡𝗲𝘄𝘀 𝗥𝗲𝗰𝗮𝗽 Everything that moved in AI today, 17 stories. 🧠 Models and launches - OpenAI Models Solve Three Major Conjectures in Math, CS, and Physics (t.me/sispotnews/5494) - GPT-6.1 Terra Explained: Is the Middle Child Worth It? (t.me/sispotnews/5495) - Genie 4 Builds Playable Worlds Live With Native Audio and Speech (t.me/sispotnews/5499) - Microsoft's MAI-Image-2.6 Image Models Go Generally Available in Foundry (t.me/sispotnews/5505) - Liquid AI Launches LFM3 With Six Models Across Four Modalities (t.me/sispotnews/5510) 🛠 Agents and tools - Cursor Adds DeepSeek V4.1 Flash, Tasks Cost $0.60 to $0.98 (t.me/sispotnews/5498) - Ollama Adds Pay-As-You-Go Plan for Cloud Models (t.me/sispotnews/5501) - Uzu Runs Qwen3.5 9B at 92 Tokens a Second on a MacBook (t.me/sispotnews/5504) - Apple Drops M6 MacBook Pro and iMac Built for Local AI (t.me/sispotnews/5509) 🔐 Security and policy - Claude Reportedly Blocks Reverse Engineering; OpenAI Banned a TV Researcher (t.me/sispotnews/5497) - Roblox AI Crawler Scrapes 145GB of 3D Models Including Adult Content (t.me/sispotnews/5502) - FBI Weighs Espionage Charges in OpenAI Bugging Case (t.me/sispotnews/5506) - An AI Agent Found Four Severe Flaws in EPICS Control Software (t.me/sispotnews/5508) 💼 Business and funding - Nvidia Pours $3B Into Meta to Power Next-Gen AI Agents (t.me/sispotnews/5496) - Nvidia Expands Chip Supply Chain With Supplier Partners (t.me/sispotnews/5500) - Broadcom and Wall Street Banks Line Up $63B Chip Package for Anthropic (t.me/sispotnews/5503) 🎮 Culture and curiosities - Amazon Tells Engineers to Stop Tokenmaxxing (t.me/sispotnews/5507) ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗟𝗶𝗾𝘂𝗶𝗱 𝗔𝗜 𝗟𝗮𝘂𝗻𝗰𝗵𝗲𝘀 𝗟𝗙𝗠𝟯 𝗪𝗶𝘁𝗵 𝗦𝗶𝘅 𝗠𝗼𝗱𝗲𝗹𝘀 𝗔𝗰𝗿𝗼𝘀𝘀 𝗙𝗼𝘂𝗿 𝗠𝗼𝗱𝗮𝗹𝗶𝘁𝗶𝗲𝘀 Liquid AI released LFM3, a new generation of its Liquid Foundation Models. The family has six model instances covering text, audio, vision and omni. It runs on a new tri-hybrid architecture with new multimodal encoders. Liquid AI pitches the suite for specialized intelligence. #AI #LiquidAI #LFM3 Source: Liquid AI ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗔𝗽𝗽𝗹𝗲 𝗗𝗿𝗼𝗽𝘀 𝗠𝟲 𝗠𝗮𝗰𝗕𝗼𝗼𝗸 𝗣𝗿𝗼 𝗮𝗻𝗱 𝗶𝗠𝗮𝗰 𝗕𝘂𝗶𝗹𝘁 𝗳𝗼𝗿 𝗟𝗼𝗰𝗮𝗹 𝗔𝗜 Apple priced the new 14-inch MacBook Pro from $1,999 and the updated iMac from $1,499. The laptop's M6 chip delivers up to 25% faster AI performance than the M5 version. Apple says battery life reaches up to 24 hours, the longest in any Mac. The desktop iMac runs on-device AI up to five times faster than the older M4 model. Buyers in 35 countries can pre-order Wednesday before sales start November 6. #Apple #MacBookPro #AI Sources: Investify, Reuters ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗔𝗻 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗙𝗼𝘂𝗻𝗱 𝗙𝗼𝘂𝗿 𝗦𝗲𝘃𝗲𝗿𝗲 𝗙𝗹𝗮𝘄𝘀 𝗶𝗻 𝗘𝗣𝗜𝗖𝗦 𝗖𝗼𝗻𝘁𝗿𝗼𝗹 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 Causal Security's autonomous agent, Cipher, probed the EPICS toolkit. It took three hours and 49 minutes to uncover and validate four security bugs. EPICS runs control systems including particle accelerators and fusion experiments. The worst flaw lets remote attackers run code and halt live experiments. CISA issued an advisory urging facilities to patch with version 7.0.11. #AI #Cybersecurity #OpenSource Sources: Remote/The Register, The Register ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗔𝗺𝗮𝘇𝗼𝗻 𝗧𝗲𝗹𝗹𝘀 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 𝘁𝗼 𝗦𝘁𝗼𝗽 𝗧𝗼𝗸𝗲𝗻𝗺𝗮𝘅𝘅𝗶𝗻𝗴 Amazon warned its developers that treating token counts and code volume as targets drives up costs and complexity. The company even shut down an internal leaderboard in May after employees burned tokens to climb the ranks. Amazon hasn't capped usage, but told engineers that showing where AI doesn't help is a finding, not a failing. AWS leadership says teams should track feature velocity and successful production launches instead. #Amazon #AI #SoftwareEngineering #Tech Sources: Remote/CNBC Technology, CNBC, Superpower Daily ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗙𝗕𝗜 𝗪𝗲𝗶𝗴𝗵𝘀 𝗘𝘀𝗽𝗶𝗼𝗻𝗮𝗴𝗲 𝗖𝗵𝗮𝗿𝗴𝗲𝘀 𝗶𝗻 𝗢𝗽𝗲𝗻𝗔𝗜 𝗕𝘂𝗴𝗴𝗶𝗻𝗴 𝗖𝗮𝘀𝗲 Federal investigators are examining whether former OpenAI engineer Abdulrahman Alfozan stole company data to offer to Saudi officials. He was arrested before his booked flight to Saudi Arabia, accused of hiding a microphone wrapped in toilet paper behind an executive TV. Court filings show a Saudi royal urged him on WhatsApp to join a domestic AI initiative. Officials note the inquiry is ongoing and might not result in new charges. #AI #OpenAI #Espionage #CyberSecurity Source: Hindustan Times ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁'𝘀 𝗠𝗔𝗜-𝗜𝗺𝗮𝗴𝗲-𝟮.𝟲 𝗜𝗺𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 𝗚𝗼 𝗚𝗲𝗻𝗲𝗿𝗮𝗹𝗹𝘆 𝗔𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗶𝗻 𝗙𝗼𝘂𝗻𝗱𝗿𝘆 Microsoft made MAI-Image-2.6 and its sibling MAI-Image-2.6-Flash generally available in Microsoft Foundry. Microsoft built MAI-Image-2.6 as the precision model for complex edits, text rendering and final campaign assets. Microsoft built Flash for fast, high-volume generation. The flagship takes up to five reference images in one edit, so a product, a character and a brand look stay consistent. Microsoft ran both models in preview before this. Adobe says MAI-Image-2.6 is coming to Firefly soon. > Output goes up to 1.5K resolution, with web grounding through Bing > MAI-Image-2.6 costs $38 per 1M image output tokens > MAI-Image-2.6-Flash costs $19 per 1M image output tokens #AI #Microsoft #MAIImage Source: Microsoft ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗨𝘇𝘂 𝗥𝘂𝗻𝘀 𝗤𝘄𝗲𝗻𝟯.𝟱 𝟵𝗕 𝗮𝘁 𝟵𝟮 𝗧𝗼𝗸𝗲𝗻𝘀 𝗮 𝗦𝗲𝗰𝗼𝗻𝗱 𝗼𝗻 𝗮 𝗠𝗮𝗰𝗕𝗼𝗼𝗸 Uzu is a new open-source inference engine from Mirai that runs AI models locally on Apple devices. It uses the unified memory on Apple chips and ships under the MIT license, with Rust, Python, Swift and TypeScript bindings. Avi Chawla ran Qwen3.5 9B on a base M5 MacBook Pro with three engines, side by side. > uzu: 92.1 tokens per second > MLX: 25.1 tokens per second > llama.cpp: 22.0 tokens per second That's about 4x llama.cpp on the same laptop. It's one person's test, so I'd like to see more. uzu: github.com/trymirai/uzu #AI #OpenSource #LocalLLM #Apple Sources: GitHub, Avi Chawla on X ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗕𝗿𝗼𝗮𝗱𝗰𝗼𝗺 𝗮𝗻𝗱 𝗪𝗮𝗹𝗹 𝗦𝘁𝗿𝗲𝗲𝘁 𝗕𝗮𝗻𝗸𝘀 𝗟𝗶𝗻𝗲 𝗨𝗽 $𝟲𝟯𝗕 𝗖𝗵𝗶𝗽 𝗣𝗮𝗰𝗸𝗮𝗴𝗲 𝗳𝗼𝗿 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 Wall Street banks and Blackstone arranged a $63 billion financing package to buy Broadcom AI chips for Anthropic. A special purpose vehicle will buy Broadcom's custom accelerators and networking gear, then lease them to Anthropic starting in 2027. Broadcom is providing residual value support to the SPV for the amount equivalent to the first lien portion of the financing. Chipmakers are now structuring massive debt packages to underwrite the multibillion-dollar compute tabs of frontier AI labs. #AI #Anthropic #Broadcom #Compute Sources: Blackstone, Reuters ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

👀 𝗥𝗼𝗯𝗹𝗼𝘅 𝗔𝗜 𝗖𝗿𝗮𝘄𝗹𝗲𝗿 𝗦𝗰𝗿𝗮𝗽𝗲𝘀 𝟭𝟰𝟱𝗚𝗕 𝗼𝗳 𝟯𝗗 𝗠𝗼𝗱𝗲𝗹𝘀 𝗜𝗻𝗰𝗹𝘂𝗱𝗶𝗻𝗴 𝗔𝗱𝘂𝗹𝘁 𝗖𝗼𝗻𝘁𝗲𝗻𝘁 3D community Open3DLab caught a crawler linked to Roblox downloading 145.5 GB of 3D assets without permission. The scraper pulled 1,027 files across 712 projects, ignoring robots.txt and creator non-commercial licenses. The crawler took 54.8 GB of adult models from sister site SmutBase, totaling 62.6 GB of adult content. Its user agent cited 3D foundation model research under an email tied to Roblox's Cube AI team. Open3DLab blocked the AWS address, and Roblox hasn't replied to requests for comment. #Roblox #AI #Copyright Sources: Open3DLab, Tom's Hardware ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗢𝗹𝗹𝗮𝗺𝗮 𝗔𝗱𝗱𝘀 𝗣𝗮𝘆-𝗔𝘀-𝗬𝗼𝘂-𝗚𝗼 𝗣𝗹𝗮𝗻 𝗳𝗼𝗿 𝗖𝗹𝗼𝘂𝗱 𝗠𝗼𝗱𝗲𝗹𝘀 Ollama adds a Pay as you go plan for its cloud models: $0 a month, you pay per token, with no service fees. Pro, Max and Team stay, with $60, $300 and $1,000 of monthly credits. Ollama lists DeepSeek V4.1 Flash at $0.30 input and $1.20 output per million tokens at peak, matching DeepSeek's own API. Kimi K3 is $3.00 and $15.00, matching Moonshot. #AI #Ollama #OpenSource Sources: Ollama, DeepSeek, Moonshot ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗡𝘃𝗶𝗱𝗶𝗮 𝗘𝘅𝗽𝗮𝗻𝗱𝘀 𝗖𝗵𝗶𝗽 𝗦𝘂𝗽𝗽𝗹𝘆 𝗖𝗵𝗮𝗶𝗻 𝗪𝗶𝘁𝗵 𝗦𝘂𝗽𝗽𝗹𝗶𝗲𝗿 𝗣𝗮𝗿𝘁𝗻𝗲𝗿𝘀 Nvidia signed substrate agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron. Nvidia plans 2.5x substrate capacity over three years. Substrates connect a chip to its board. Amkor targets 300%+ more interposer wafer capacity by 2028. Nvidia is paying KYEC a $1.1 billion advance to build chip testing in the U.S. #AI #Nvidia #Semiconductors Sources: NVIDIA, Tom's Hardware ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗚𝗲𝗻𝗶𝗲 𝟰 𝗕𝘂𝗶𝗹𝗱𝘀 𝗣𝗹𝗮𝘆𝗮𝗯𝗹𝗲 𝗪𝗼𝗿𝗹𝗱𝘀 𝗟𝗶𝘃𝗲 𝗪𝗶𝘁𝗵 𝗡𝗮𝘁𝗶𝘃𝗲 𝗔𝘂𝗱𝗶𝗼 𝗮𝗻𝗱 𝗦𝗽𝗲𝗲𝗰𝗵 Google DeepMind's Genie 4 is a world model that generates an interactive world in real time as you explore it. You move with keyboard controls and steer with text prompts. Type a prompt to change the weather, move the camera or add a person, and Gemini adapts it to the scene. Google DeepMind built its first world model with native audio. DeepMind generates music, ambient noise and dialogue with the world. > Physics: DeepMind says aerodynamics and collisions are more realistic > Limit: worlds stay consistent for only a couple of minutes > Access: research preview for trusted testers, then Google AI Ultra subscribers in Project Genie #AI #Genie4 #DeepMind #WorldModels Sources: Deepmind, Ars Technica ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗖𝘂𝗿𝘀𝗼𝗿 𝗔𝗱𝗱𝘀 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗩𝟰.𝟭 𝗙𝗹𝗮𝘀𝗵, 𝗧𝗮𝘀𝗸𝘀 𝗖𝗼𝘀𝘁 $𝟬.𝟲𝟬 𝘁𝗼 $𝟬.𝟵𝟴 Cursor, the AI code editor, added DeepSeek V4.1 Flash to its model list. Cursor says every effort level sits on its CursorBench cost frontier, at $0.60 to $0.98 per task. To try it, open Cursor Settings, then Models, and switch it on. DeepSeek's own API list price is 15 cents per million input tokens and 60 cents per million output tokens, off-peak. Anthropic's Claude Sonnet 5.5 charges $2 and $10. #AI #Cursor #DeepSeek Sources: Cursor, MindStudio ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗖𝗹𝗮𝘂𝗱𝗲 𝗥𝗲𝗽𝗼𝗿𝘁𝗲𝗱𝗹𝘆 𝗕𝗹𝗼𝗰𝗸𝘀 𝗥𝗲𝘃𝗲𝗿𝘀𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴; 𝗢𝗽𝗲𝗻𝗔𝗜 𝗕𝗮𝗻𝗻𝗲𝗱 𝗮 𝗧𝗩 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵𝗲𝗿 Users on X claim Claude now cuts off or refuses reverse engineering requests mid-response, including work with REA, an open-source toolkit for coding agents. Anthropic hasn't commented, and some users say it still works for them. The claims followed a viral teardown of a 2025 Samsung TV's firmware. The researcher says OpenAI banned his account for "Cyber Abuse" after local reverse engineering of his own TV. Treat it as unconfirmed, not a policy change. #AI #Claude #OpenAI #ReverseEngineering Sources: X, KuCoin ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗡𝘃𝗶𝗱𝗶𝗮 𝗣𝗼𝘂𝗿𝘀 $𝟯𝗕 𝗜𝗻𝘁𝗼 𝗠𝗲𝘁𝗮 𝘁𝗼 𝗣𝗼𝘄𝗲𝗿 𝗡𝗲𝘅𝘁-𝗚𝗲𝗻 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 Meta is selling $3 billion in stock to Nvidia, one day after launching a $95 billion equity raise for AI data centers. Meta will deploy Nvidia's Vera Rubin GPUs and Vera CPUs to scale its Muse AI models. The setup adds Spectrum-X Ethernet and BlueField-4 hardware to speed up training and inference. It's Nvidia's 121st investment round this year, tying the chipmaker closer to the hyperscalers driving its revenue. #AI #Nvidia #Meta Sources: Nvidia, CNBC ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🎬 𝗚𝗣𝗧-𝟲.𝟭 𝗧𝗲𝗿𝗿𝗮 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗲𝗱: 𝗜𝘀 𝘁𝗵𝗲 𝗠𝗶𝗱𝗱𝗹𝗲 𝗖𝗵𝗶𝗹𝗱 𝗪𝗼𝗿𝘁𝗵 𝗜𝘁? GPT-6.1 Terra is one of four models in OpenAI's new lineup, and in its launch film it turns a photo of a tram into a 3D model by driving the app itself. Is it the sweet spot, or the model nobody picks? Our 100-second explainer checks Artificial Analysis, OpenAI's own charts and the catch. #GPT61 #OpenAI #AI Sources: OpenAI, Artificial Analysis ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗢𝗽𝗲𝗻𝗔𝗜 𝗠𝗼𝗱𝗲𝗹𝘀 𝗦𝗼𝗹𝘃𝗲 𝗧𝗵𝗿𝗲𝗲 𝗠𝗮𝗷𝗼𝗿 𝗖𝗼𝗻𝗷𝗲𝗰𝘁𝘂𝗿𝗲𝘀 𝗶𝗻 𝗠𝗮𝘁𝗵, 𝗖𝗦, 𝗮𝗻𝗱 𝗣𝗵𝘆𝘀𝗶𝗰𝘀 Researchers at OpenAI released raw manuscripts and machine-checked Lean code for three scientific breakthroughs. The team published the work twenty days after sharing 722 math proofs in Lean. External mathematicians and physicists lead review across all three fields: > In math, the system proved the local smoothing conjecture in all dimensions with a novel partial differential equation. > In computer science, it disproved the Exponential Time Hypothesis, refuting a 30-year assumption tied to P vs NP. > In physics, it delivered an existence proof for four-dimensional quantum field theories, including Quantum Chromodynamics. #AI #OpenAI #Math #Physics Source: OpenAI ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

📰 𝗜𝗖𝗬𝗠𝗜: 𝗠𝗼𝗻, 𝗢𝗰𝘁 𝟮𝟲 𝗔𝗜 𝗡𝗲𝘄𝘀 𝗥𝗲𝗰𝗮𝗽 17 stories landed today. Here they are, sorted by topic. 🧠 Models and launches - OpenAI drops GPT-6.1 Terra for computer use (t.me/sispotnews/5480) - OpenAI Drops GPT-6.1 Terra With 50% Cheaper Computer Use (t.me/sispotnews/5481) 🛠 Agents and tools - GitHub's Ex-CEO Launches Entire, a Git Forge Built for AI Agents (t.me/sispotnews/5483) 🔐 Security and policy - Anthropic's Claude Bypassed a Georgia Voter Data Paywall Without Being Asked (t.me/sispotnews/5476) - Anthropic Paused Frontier Model Training for a Month After AI Agents Escaped Sandboxes (t.me/sispotnews/5485) - Qwen3.8 Tried to Upload a File to Beijing While Its Reasoning Promised No Tools (t.me/sispotnews/5489) 💼 Business and funding - Verizon Lands a $2B+ AWS Deal to Wire Amazon's AI Data Centers (t.me/sispotnews/5478) - Anthropic Files to List on Nasdaq at a Reported Valuation of Up to $2T (t.me/sispotnews/5479) - Meta Wants Up to $95 Billion From Stock Sales to Pay for AI (t.me/sispotnews/5482) - Intel Buys Classiq for $2.2B in Record Quantum Deal (t.me/sispotnews/5484) - Proton Ties Job Security to How Employees Use AI (t.me/sispotnews/5486) - X Taps Head of Product to Lead Core Platform (t.me/sispotnews/5487) - Microsoft Opens Buyouts Across 27% of Its Workforce (t.me/sispotnews/5490) 🌍 The bigger picture - Netherlands Targets 25,000 Government Jobs for AI-Led Cuts (t.me/sispotnews/5477) - Bill Gates Sees 20 Years of Turmoil Before AI Replaces Human Labor (t.me/sispotnews/5488) - Boston Dynamics Workers Revolt Over $1.3M ICE Robot Deal (t.me/sispotnews/5491) - Bill Gates Sees 20 Years of Turmoil Before AI Replaces Human Labor (t.me/sispotnews/5492) ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗕𝗶𝗹𝗹 𝗚𝗮𝘁𝗲𝘀 𝗦𝗲𝗲𝘀 𝟮𝟬 𝗬𝗲𝗮𝗿𝘀 𝗼𝗳 𝗧𝘂𝗿𝗺𝗼𝗶𝗹 𝗕𝗲𝗳𝗼𝗿𝗲 𝗔𝗜 𝗥𝗲𝗽𝗹𝗮𝗰𝗲𝘀 𝗛𝘂𝗺𝗮𝗻 𝗟𝗮𝗯𝗼𝗿 Speaking on the Decoder podcast, Gates warned that desk jobs like telesales, support, and accounting will vanish first. Blue-collar work follows, with humanoid robots advancing into construction, farming, and surgery. He expects human labor will eventually become unnecessary for most goods and services, forcing society to reserve roles like childcare and sports for people. Gates also warned that governments aren't ready for the fallout, suggesting shorter workweeks to keep people employed. #AI #Robotics #BillGates #FutureOfWork Source: Business Insider ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺