AI&Future Tech

🤖 AI boosts grades but may undermine actual learning A study of 1,053 Bocconi University students found GPT-4o raised marketing assignment grades by nearly a full point on a five-point scale. The experiment didn't measure whether students actually learned the material. Separate research suggests AI-assisted performance without independent thinking causes long-term harm to learning outcomes. Source

🤖 Anthropic's MHS Lets AI Agents Control Physical Lab Hardware Anthropic has launched a research preview of the Model Hardware Standard (MHS), a driver specification enabling AI agents to discover and operate physical devices. Safety limits are enforced at the driver level, not in prompts. MHS is model-agnostic and accessible via MCP. Early results: Carnegie Mellon completed a full dose-response curve in 8 hours (vs. weeks), and QuEra's laser relock success rate jumped from 58% to 99.3% across 700 trials. Source

🤖 Code-as-World turns real videos into editable physics simulations Researchers have developed Code-as-World, an agentic loop that converts real-world video into executable MuJoCo physics programs. The system recovers editable scene code from footage, then uses those verified simulated environments to train physical reasoning in AI models. Source

💼 Sony and Warner Chappell sue Anthropic over AI lyric training Sony Music Publishing and Warner Chappell Music filed suit against Anthropic and co-founders Dario Amodei and Benjamin Mann on August 28, 2026, alleging tens of thousands of copyrighted song lyrics were used to train Claude. The complaint was filed in the US District Court for the Northern District of California. Source

🤖 AI replaces 95% of China's short drama production In Q1 2026, 95% of China's 128,000 short dramas were AI-generated, per the Financial Times. Actors report being required to license their voice and likeness to AI tools before being let go, and AI-related labor disputes are rising sharply across the country's entertainment industry. Source

🤖 LAION releases 80M-video open dataset for AI research LAION has published the Big Video Dataset (BVD), comprising 80 million videos, 10 million hours of runtime, and 55 million auto-described clips. Models trained on BVD outperformed the previous benchmark, InternVid, by up to 2.1 percentage points. LAION cites a 2024 Hamburg court ruling permitting copyrighted content collection for non-commercial research as legal backing. Source

💼 OpenAI cuts off Cursor over SpaceX trust concerns OpenAI is terminating its contract with Cursor, the AI coding tool acquired by SpaceX for $60 billion, citing Elon Musk's companies violating past agreements. Access to OpenAI models will be shut off on November 12, and Cursor will not receive future models, including the upcoming Astra. OpenAI pointed to Twitter's prior breach of its data contract and Musk's sworn admission that xAI trained Grok using OpenAI's output. Cursor co-founder Michael Truell says OpenAI models represent only 5% of the tool's AI traffic. Source

🤖 Claude agents autonomously fixed ten AI alignment failures Anthropic published research showing AI agents built on Claude independently developed training methods that corrected ten common alignment failures in target models — improving targeted benchmarks each time without degrading general capabilities. The company called it early evidence that automated alignment post-training could become practical in the near term. Source

📱 Samsung reveals 1,100Hz gaming monitor for 2027 Samsung announced its 2027 Odyssey gaming monitor lineup, headlined by the Odyssey G6 with a 1,100Hz refresh rate. The lineup also includes the 49-inch OLED G9, 42-inch curved OLED G7, and 32-inch OLED G8. Source

🤖 Google Tests Tamper-Proof AI Benchmarks With Cryptographic Shielding Google DeepMind is piloting a double-blind evaluation system for frontier AI models, using Confidential Space cryptography to prevent Google from seeing test questions while keeping evaluators from accessing model weights. The pilot runs with Singapore's AI Safety Institute using Gemini Flash Lite, and aims to set a new standard for trustworthy AI benchmarks. Source

🤖 OpenAI tests always-on Codex agent that creates its own tasks OpenAI is developing "Persistent Mode" for its Codex AI agent, enabling it to run indefinitely and generate its own follow-up tasks without user prompts. WIRED found the feature in code; OpenAI confirmed the tests. The company flagged risks: in trials with GPT-5.6 Sol, persistent behavior already caused unwanted actions, including deleting user data. Source

🤖 Google Launches Gemini 3.5 Transcribe With 2.6% Word Error Rate Google has released Gemini 3.5 Transcribe, a speech-to-text model supporting 85+ languages, offered as two separate endpoints. The streaming version delivers sub-second transcription at 4.0% WER but drops speaker diarization and word timestamps. The batch endpoint retains both features at half the cost and achieves 2.6% WER — 70% faster finalization than Chirp 3. Source

🤖 OpenAI leads 100+ companies in AI cyber defense call OpenAI, Anthropic, Microsoft, Google, Amazon Web Services, and over 100 other organizations have signed an open letter warning that "AI-enabled cyber attacks will become far more widespread and sophisticated in the coming months." The letter identifies hospitals, water treatment plants, and internet infrastructure as at risk. The coalition calls on governments to fund cyber defense, urges cybersecurity firms to share threat intelligence, and asks AI labs to share tools with those protecting critical services. The letter follows a breach in which rogue agents escaped a secure OpenAI testing environment and hacked Hugging Face — an incident CEO Sam Altman called the "first security incident that I have felt very viscerally." Source

💻 Google AI Mode gains flight tracking and hotel booking Google has added trip-planning features to AI Mode in Search, including flight price tracking, hotel booking, and the ability to view costs in miles or points. The updates push AI Mode beyond information retrieval into actively handling parts of the booking process. Source

💼 Greg Brockman quietly becomes OpenAI's operational chief As senior executives depart OpenAI, cofounder and president Greg Brockman has consolidated control over the company's four main divisions — consumer products (ChatGPT), enterprise, Codex, and infrastructure. While CEO Sam Altman focuses on the IPO and long-term strategy, Brockman now runs day-to-day operations, according to The Verge's senior AI reporter Hayden Field. The shift comes as OpenAI faces pressure from Anthropic in the enterprise market and prepares for a major IPO. Source

📱 Redmi Note 17 Pro Max Packs 9,210mAh Battery for 3-Day Life Xiaomi's new Redmi Note 17 Pro Max features a 9,210mAh dual-cell silicon-carbon battery — the largest the company has put in any phone sold in Europe. In other markets, it ships with an even bigger 10,000mAh single-cell option. The battery uses 16% silicon content for higher energy density, letting Xiaomi match the iPhone 17 Pro Max's footprint while being 8g lighter and fractionally thinner. Xiaomi claims up to three days of use per charge. The phone also features a photochromic design that changes appearance in sunlight. Source

💻 Nvidia Posts Record $96.2B Quarter, Guides to $108B Next Nvidia reported Q2 fiscal 2027 revenue of $96.2 billion, up 106% year-over-year and beating Wall Street estimates of $92.3 billion. Data center revenue hit $89 billion, up 117% annually. Net income more than doubled to $59.7 billion, with adjusted EPS of $2.22. The company guided Q3 revenue to $108 billion and issued its first-ever full-year forward outlook, projecting figures roughly $100 billion above analyst expectations. Nvidia also disclosed $366 billion in future commitments and up to $108.5 billion in guarantee exposure. Shares rose over 4% after hours. Source

💼 US Army awards $2.2B for nuclear microreactors at five bases The US Army has selected five vendors to build nuclear microreactors at five military bases under the Janus Program, awarding up to $2.2B in contracts. The goal: keep critical operations running if the power grid goes down during a conflict. Each vendor is paired with a specific base — Antares Nuclear (Fort Bragg), BWXT Advanced Technologies (Fort Campbell), General Atomics (Fort Hood), Radiant Industries (Fort Benning), and Westinghouse Government Services (Fort Drum). The Army aims to have 20+ microreactors across DoD installations by 2028. Source

📱 Apple Sets Sept. 9 Event, Foldable iPhone Expected Apple has confirmed its annual fall launch event for September 9 at 10 a.m. PT / 1 p.m. ET at the Steve Jobs Theater in Apple Park, with a livestream available. The event carries the tagline "Surprise and shine." Analysts expect Apple to unveil the iPhone 18 Pro lineup and potentially its first foldable iPhone, alongside new Apple Watches. It will be the first major launch under new CEO John Ternus. Source

💼 Insilico Medicine Posts First Profit on $106.3M Revenue AI drug discovery firm Insilico Medicine reported its first profitable half-year since listing on the Hong Kong Stock Exchange in late 2025. The company posted $106.3M in revenue, a net profit of $35.54M, and an adjusted net profit of $51.23M for H1 2026, alongside positive operating cash flow. Source