🌍 Type "make it rain" while you walk, and Google's Genie 4 rebuilds the world around you
Google DeepMind's Genie 4 generates a world you can walk through in your browser, and now you can change it mid-stroll: shift the weather, drop in a character or redirect the action. For the first time it also makes its own live sound, from wind in the trees to characters talking back. Output is real-time 720p at up to 24 frames per second. It's rolling out to eligible Google AI Ultra subscribers in Project Genie; Pro users can join a waitlist.
⭐ Video games may one day be typed, not coded. Google admits details still drift in long sessions.
⚡ Google's new Gemini 4 Neon is now #1 on a coding-agent ranking, at $3 per million tokens
Google just launched Gemini 4 Neon, a fast model built to finish long coding, office and science jobs end to end. It tops the Artificial Analysis Coding Agent Index (58.6) and ranks #3 overall on its Intelligence Index, just behind bigger, slower models. It even won Vending-Bench 2, a simulation of running a business, with $16,843. Price: $3 in / $15 out per million tokens, with a 1 million-token context window. It's rolling out now in the Gemini API, Antigravity and the Gemini app for Pro and Ultra users.
⭐ Near-top AI is getting fast and cheap enough to hand it a whole job, not just a question.
🚧 Waymo just pulled every driverless taxi off US freeways
Waymo has temporarily stopped its robotaxis from driving on freeways in all its US markets. It follows California's ban, which the DMV called an "imminent hazard to public safety" after Waymo cars repeatedly drove into active construction zones, including one on US Route 101 south of San Francisco. No injuries or collisions were reported. City-street rides continue in its 15 US cities.
⭐ Robotaxis are winning the city streets, but highways at speed are proving a much harder test of trust.
🤖 Microsoft's new Autopilot agent works around the clock, even while you sleep
At Ignite, Microsoft began rolling out Autopilot, an always-on agent with its own identity, memory and mailbox that you can @mention in Teams and Outlook like a coworker. It watches channels, chases emails, runs recurring work and can manage a whole supplier review on its own, checking with you before key actions. Microsoft also made Word, Excel and PowerPoint work right inside the Copilot app, at no extra usage cost for licensed users. Autopilot starts with Frontier customers and is billed by usage.
⭐ Office work is flipping: the AI agent drives, and the apps you know become tools it uses.
🚀 OpenAI's new GPT-6.2 Astra beats its old flagship for under a third of the cost
OpenAI just released GPT-6.2 Astra, its model for the hardest work, and GPT-6.2 Sol for everyday tasks. Astra scored 5.2 points higher than GPT-6 Astra on Agents' Last Exam, a test of multistep professional work, at less than a third of the cost per task. Sol came within half a point of Astra on a real-codebase coding test at about a fifth of the cost. Both are live now in ChatGPT Work, Codex and the API, and Astra's API price drops 20% to $8/$40 per million tokens.
⭐ Frontier AI keeps getting cheaper fast: the same budget now buys noticeably smarter agents for coding, slides and spreadsheets.
🗣️ AI Tip: Stop yelling "NEVER" at your AI. Tell it why instead
Rules in all caps give the model a command but no goal. Anthropic's prompting guide says adding the reason behind an instruction helps Claude deliver more targeted answers, because it can generalize from the explanation.
Less effective: NEVER use ellipses
More effective: Your response will be read aloud by a text-to-speech engine, so never use ellipses, since it won't know how to pronounce them.
Now the model also knows to skip other things a voice can't read, not just "...".
Same trick works for any rule: "Keep it under 100 words" becomes "Keep it under 100 words, since it's shown in a phone notification."
🏥 Sheba Medical Center put a second robot brand on the same AI dispatcher
Gix Internet said on November 17 that an Ottonomy delivery robot is now in service at Israel's largest hospital, beside robots from other makers. The deliverz.ai platform assigns the jobs, monitors them, and also lines up human couriers. Sheba already uses it for chemotherapy drugs, and has extended it to blood and bone marrow. Gix gave no dollar figure for this deployment. The company describes the robot as Level 4 autonomy.
⭐ The news is one control layer running more than one robot brand.
🧪 A 27-billion-parameter model practiced in simulation, then finished 18 of 30 real robot trials
Researchers trained the open model Qwen3.8-27B with reinforcement learning and no human demonstrations. The setup is called PEAR. On a real two-arm robot it never saw in training, the final model succeeded 18 times out of 30. The base model succeeded zero times. On the LIBERO benchmark, the success rate rose from 1.6% to 74.4%. The paper was posted to arXiv on November 13.
⭐ Sixty percent transferred out of the simulator. Twelve of the thirty real tries still failed.
💰 Genesis AI raised $500 million, at a $3.5 billion valuation
Khosla Ventures, Premji Invest and Eclipse led the Series A, with NVIDIA, AMD Ventures and Bpifrance among the backers, the company said on November 17. The cash is for Eno, a robot on a wheeled base whose body folds and stretches instead of walking on two legs. Genesis told TechCrunch it has built 30 to 40 Enos and is piloting them with a handful of customers. A contractor is due to start mass production in the US next year.
⭐ The valuation is for a robot still in customer pilots, not a fleet already on a factory floor.
📦 Humanoid robots are sorting parcels at 11 China Post hubs
ROBOTERA said on November 17 that the rollout covers 11 hubs in eight Guangdong cities, including Guangzhou, Shenzhen and Dongguan. The Guangzhou pilot started in March 2026. The company says the robots put more than 1,500 parcels an hour onto the sorting line, about the pace of an experienced worker, using a five-finger hand it built itself.
⭐ Those throughput numbers are the company's. The robots are on a live postal line.
🤖 Atlas gets its first chip-plant trial, starting in early 2027
Boston Dynamics and GlobalFoundries said on November 17 that the chipmaker will be the first customer outside the auto industry to test the humanoid. Spot, the four-legged robot, is also expanding across fabs in New York, Vermont, Germany and Singapore. Two Spots in Singapore have already traveled nearly 900 kilometers and completed more than 90,000 inspections since April, with a 99% mission success rate, the companies said.
⭐ Atlas is not running a chip line yet. This is a test slot, and the first one outside car factories.
🚗 California barred Waymo from every freeway in the state
The DMV order took effect on November 11 and became public on November 17. On November 5, three driverless Waymos entered an active construction closure on US-101 near Burlingame. The next day, another vehicle met California Highway Patrol on Highway 280 in San Francisco. The department called the freeway operation an imminent hazard to public safety. Waymo then paused commercial freeway rides in every market where it runs them. Pickups on city streets are still going.
⭐ The ban covers California freeways. Waymo turned the highway service off everywhere else on its own.
🧠 Qwen4-Max leaves preview with 2.4 trillion parameters
Alibaba has launched its new flagship with native vision and a 1-million-token context window. In its own Work360 tests, average task cost fell from $4.43 to $1.67 versus Qwen3.8-Max. Its demos span game creation, manufacturing and scientific workflows; smaller Qwen4 models are still to come.
⭐ The big change: a full release with published architecture and cost results.
⚡ AI Tip: Editing a file with AI? Hand it the old version as a "prediction"
Most of a refactored file stays the same. OpenAI's Predicted Outputs let you pass the current text in the prediction parameter of Chat Completions, so the model can reuse matching tokens and return the whole file with lower latency.
Example: ask GPT-4.1 to rename username to email in a JS class, and send the original code as {"type": "content", "content": code}.
Watch the fine print: • Works with GPT-4o, GPT-4o mini and the GPT-4.1 family • Prediction tokens that don't make it into the answer are still billed as output; check rejected_prediction_tokens in usage • No function calling, n > 1, logprobs or max_completion_tokens
🌐 Repo of the Day: Google Chrome wants to stop your coding agent from writing 2015-era web code
Modern Web Guidance is a set of skills that feeds web platform expertise, best practices and browser compatibility data straight into your coding agent. The idea: LLMs learned from mountains of legacy code, so they often write bloated JavaScript for things the browser now does natively.
Who it's for: web developers who build with AI coding agents. One command installs it: npx modern-web-guidance@latest install
Standout feature: it covers 250 modern web features, from container queries and View Transitions to the Popover API and built-in AI APIs, in token-efficient guides pruned with evals. A legacy-scan command can even flag old patterns in your codebase to modernize. It is stewarded by the Chrome, Edge, Firefox and MDN teams.
🖥️ A Mac browser for coding agents keeps every tab in one window and never steals the keyboard
Agweb, posted on Hacker News today, is a WebKit browser for macOS 26. Agents drive it from the terminal while the window stays in the background. Its author measured five idle pages at 590 MB, next to 730 MB for headless Chrome and 1.6 GB for Playwright's Chrome, and those pages sat at 0% CPU on the author's Mac. Several agents can share one login profile at the same time. The public repo has 45 stars.
⭐ The annoying part of a browser agent is the window jumping in front of you. This one refuses.
✈️ The maker of the FAA's AI traffic tool says SMART did not ground US flights
Air Space Intelligence posted that the nationwide ground stop "is not based on or a result of the SMART system," any ASI product, or its link to air traffic control. SMART is the FAA's planning tool. The Transportation Department says ASI's contract to build it is $875 million. It reads about 200 data streams, from weather to controller staffing, and suggests routes. People still approve every change. In the Washington trial, the FAA said it rerouted about 180 flights, saving about 10 minutes and nearly 80 gallons of fuel each.
⭐ The AI proposes a route. A person still has to say yes.
🖥️ One GPU cannot hold GLM-5.3, so Nvidia split the 753-billion-parameter model across two machines
Nvidia's new guide runs Z.ai's GLM-5.3, quantized to NVFP4, on two DGX Stations linked by ConnectX-8. The checkpoint is about 464 GB. One GB300 GPU has 252 GB of memory, so each machine keeps about 218 GiB of the weights and they answer through one API. On Nvidia's coding test, output peaked at 607.8 tokens a second with 16 requests at once, then fell to 385 when 32 ran together.
⭐ The fast number is a crowd result. One person still gets about 115 tokens a second.
🧠 Changing the plan, not the model, lifted a robot world model's success from 22% to 85%
Researcher Haddy Alchaer ran 600 experiments on LeWM, a JEPA world model from a team that includes Yann LeCun. As published, it succeeded on 22% of simulated tasks where a robot must push a T-shaped block to a goal 75 steps away. A new subgoal every 25 steps raised that to 80%, and a memory of past demonstrations with a cost on every jump between them reached 85%. On paths the model was trained for, its own predictions scored 78%, and the simulator's real physics scored 77%. No weights were retrained.
⭐ The useful fix was a shorter plan, not a smarter predictor.
💳 Gemini can now read your Google Wallet and chart the spending in the chat
Google is rolling this out to every Gemini app user in the US, on the web, Android, and iOS. Opt in, tag @Google Wallet, and it can chart last quarter's cash flow, flag subscriptions that rose in price this year, or build a holiday budget from last year's gifts and travel. Bank and card accounts can be linked through Plaid. Gemini can read the accounts, and it cannot move, manage, or invest the money.
⭐ It can see where the money went. It still cannot touch the account.
#AI
GOOGUS+1,87%
Войдите, чтобы посмотреть больше материала
Присоединяйтесь к пользователям криптовалют по всему миру на Binance Square
⚡️ Получайте новейшую и полезную информацию о криптоактивах.
💬 Нам доверяет крупнейшая в мире криптобиржа.
👍 Получите достоверные аналитические данные от верифицированных создателей контента.