AI that fits your day.

A Wired report, originally published by High Country News, describes cases in which hikers used chatbots to plan outings and ended up in dangerous situations. Three young men on Mount Shasta in California said they partly relied on Gemini; they reached the summit late, descended in the dark, and needed rescue. In another case, a visitor camped near a “No Camping” sign while claiming AI had said she could. A teenager rescued from a ledge in British Columbia had used Claude, although he had also consulted Google Maps and AllTrails.

The article cites experts who see specific risks in chatbots’ persuasive but incorrect answers. Chris Flathmann of Clemson University calls the possibility that users stop thinking critically when treating AI as an expert “cognitive surrender.” It remains unclear whether increased AI use will lead to more accidents and rescues in remote areas.

Sources: Wired

Read the original story ↗

The Verge’s author paid $50 for a 1.75-inch touchscreen computer, the Waveshare ESP32-S3 1.75C, to build a device resembling the Muse Charm. Meta officially sanctions the model as one of its suggestions for creating a “Muse Gadget,” using the company’s open-source code and tokens to connect directly to the Muse cloud. In its initial setup, the device had slow responses, no spoken output, limited on-screen text and no automatic discovery of smart-home devices.

With sessions of Claude Code, the author built a conversation-history page, added an ElevenLabs voice, remote camera control and a live phone-camera preview, as well as new avatars and a Tamagotchi-style minigame involving hunger, boredom, dirt, food, play, sleep and bathing. The article also notes that the device lacks cameras and a 5G cellular radio, features planned for the Muse Charm.

Sources: The Verge

Read the original story ↗

Cisco Talos researchers identified Volta-Autodoxer, a new strain of infostealer malware that gathers credentials, tokens, cookies, payment cards, crypto-wallet data, and active sessions from services including Discord, Telegram, Steam, and Roblox. The malware also searches central accounts for deeper information, including UberEats and DoorDash receipts and order histories, to build a profile of the victim.

According to Wired, the code sends this material through OpenRouter to AI models that produce a structured assessment, including an executive summary, the target’s habits, estimated wealth, and operational security practices. Claude Opus 4.6 is the default model, with Kimi K2 and Gemini 2.5 Flash as fallbacks if the activity is blocked. Cisco Talos says it did not track whether Volta has been used in real-world attacks.

Sources: Wired

Read the original story ↗

A research and development center near Shanghai brought together more than 100 companies for a robot “Carnival” that drew families during a holiday. Visitors watched quadruped robots climb stairs, remotely controlled machines duel with water jets, and humanoids perform martial arts and acrobatics; there were also models designed for tasks such as carrying heavy loads, inspecting pipes and making coffee. Children could buy dog-shaped mini robots and interact directly with the machines.

According to MIT Technology Review Brasil, China made nearly 90% of the more than 13,000 humanoid robots delivered worldwide last year. Embodied AI is part of the country’s latest five-year plan, while Chinese companies continue promoting the machines in shopping malls, tourist attractions and sporting events. Yang Duan, a product manager at DexForce, said humanoids could reduce physical work and repetitive tasks, although high prices, short battery life and safety challenges still limit the technology.

Sources: MIT Technology Review Brasil

Read the original story ↗

Apple announced on its Chinese newsroom website that it will bring Apple Intelligence to China with iOS 27.3, The Verge reported, citing MacRumors. The features will first be available in an iOS 27.3 beta scheduled for November 4, while the full version is planned for December.

The release will include features such as notification summaries, real-time translation, rewriting and proofreading tools, visual intelligence, and object removal in photos, but not the new AI-powered Siri. Apple also says it worked with Ailbaba to add support for the Qwen chatbot in some experiences. Apple Intelligence was approved in China in July, and Reuters reported in August that Apple had trained an LLM specifically for the country.

Sources: The Verge

Read the original story ↗

According to Ars Technica, Dartmouth provost Schnell resigned after a scandal involving artificial intelligence use. President Sian Leah Beilock said the university should continue expanding its use of AI, but that its leaders must model transparency and human judgment—particularly by disclosing how they incorporate these tools into their work.

In a September 2026 paper about detecting AI-generated text, Schnell described using GPT-5.6 Pro, GPT-5.6 Sol and Anthropic Claude Fable 5.1 to structure ideas, produce drafts, edit language and critique the manuscript. He said he chose the topic and direction, checked sources, rewrote most of the text and took full responsibility for the result. Schnell rejected calling AI a co-author, arguing that contribution and authorship are different roles.

Sources: Ars Technica

Read the original story ↗

Unitree presented the H2-X humanoid as a “bodyguard” robot capable of intervening in dangerous situations and helping in emergencies. In the promotional video, it confronts a man with a knife, lifts a car trapping a person underneath, strikes another vehicle’s door and pulls a mannequin out of the path of a car. According to Reuters, the company rapidly expanded its humanoid lineup to explore applications in surveillance, security and rescue.

Interesting Engineering says technical data about the H2-X is still lacking and that it is unclear how the maneuvers were performed. Unitree has not announced a price or commercialization date. The video also shows the robot throwing the mannequin to the ground while trying to save it, raising questions about the risks of using a machine with so much physical strength in real-world situations.

Sources: g1

Read the original story ↗

OpenAI announced that, in the coming weeks, it will further limit the practical information about weapons and security systems provided by ChatGPT to teens, even when questions appear harmless. The company says it has improved its age-estimation system to direct users identified as under 18 to the teen-focused version.

ChatGPT may also automatically end conversations showing signs of fixation on violence and, if the behavior recurs, temporarily suspend the account and notify parents whose accounts are linked. For all users, OpenAI will reduce the level of detail about past mass attacks, including the weapons used and how they were obtained. The measures were announced amid concerns about teen safety; according to Common Sense Media, alerts sent to parents about conversations involving suicide, self-harm or eating disorders often arrive too late or are not sent.

Sources: g1

Read the original story ↗

Zest launched a personal AI concierge that works through text messages. Users can request recommendations by cuisine, occasion, dietary restrictions or neighborhood, ask questions about restaurants, receive menus and check available reservation times through services such as OpenTable, Resy and SevenRooms. The concierge can be reached at (212) 470-6067; it can also be added to group chats, where it combines participants’ tastes, runs a vote and books the selected table without leaving the conversation.

The company uses food-and-drink transaction data imported through Plaid to create personalized recommendations, and CEO Mario Gomez-Hall says this lets the agent start from users’ actual history instead of starting from scratch. In TechCrunch’s tests, the tool offered suggestions matched to preferences, time of day, party size and location. Zest says bookings and food orders will be revenue sources and is considering expanding the service to businesses, shopping and other categories.

Sources: TechCrunch

Read the original story ↗

Starbucks will explore experiences that recognize participating customers as they arrive at the drive-thru, show personalized digital menus and make payment easier, CEO Brian Niccol said. There is no timetable yet for the menus; Niccol said the initiative is still at an early stage and that the company plans to test, learn and scale what works. Starting in 2027, the chain will begin testing background AI models to help baristas enter orders more accurately and efficiently.

The company also plans to integrate AI into search and discovery in the Starbucks app and may offer conversational and agentic ordering in the future. Starbucks already allows orders through ChatGPT and has rolled out Green Dot Assist, an AI tool for employees. The plan resembles tests by Taco Bell and McDonald’s to adapt digital menus or recognize repeat customers at the drive-thru.

Sources: The Verge

Read the original story ↗

Amazon confirmed to Tecnoblog that it has extended the free early-access period for Alexa+ in Brazil until at least February 2027. The previous deadline was at least October 2026. The assistant uses generative AI for more contextual conversations, calendar management and weather questions, but may take longer to respond to complex commands.

Alexa+ will be included with Amazon Prime, starting at R$ 19.90 per month, or offered separately for R$ 99 per month. For now, Prime customers must join the waitlist and have a compatible device, such as the Echo Show 5. Amazon told Tecnoblog that it has not yet set a general-release date and will notify customers before the transition.

Sources: Tecnoblog

Read the original story ↗

David Durman, a University of Kentucky administrator and nontheistic humanist, secured a religious exemption allowing him not to use artificial intelligence at work, TechCrunch reported. With support from the American Humanist Association (AHA), he argued that reliance on AI could harm people’s ability to think and act morally, threaten human creativity and create environmental costs. The AHA says Durman is the first humanist to publicly secure such an accommodation; he was inspired by the case of Erin Maus, a Unitarian Universalist software engineer who obtained similar protections.

University spokesperson Jay Blanton told TechCrunch that requests for religious accommodations are reviewed individually, taking job duties and the circumstances of the request into account. In this case, the university determined that the accommodation could be provided without preventing Durman from performing his essential duties, since his administrative role did not require AI use. The university may review the accommodation as technology and job responsibilities change.

Sources: TechCrunch

Read the original story ↗

Meta has quietly launched Style Sketch, an AI fashion app that connects to Instagram to create a virtual closet from photos users have already posted. Users can also add items manually by taking photos, recording short videos or uploading images from their camera roll. The app suggests combinations using the available clothes and, according to its description, lets users try them on a realistic model matched to their body shape and skin tone.

Style Sketch can also create weather-based looks and day-by-day outfit plans for trips based on the destination, trip length and planned activities. The app is available on iOS and Android in Indonesia, Singapore and Thailand. Meta did not respond to TechCrunch’s request for comment, and it is unclear whether the app will launch more widely.

Sources: TechCrunch

Read the original story ↗

Google says Gemini now lets users generate, preview, and edit PDFs, Google Docs, Google Sheets, presentations, Markdown files, TXT files, and code directly within its interface, without leaving the service. The company says the feature is available to all users, in every language in which Gemini is offered and globally, including free accounts and paid plans, according to Tecnoblog.

PDF editing and previewing work on the web and in Gemini’s mobile apps; the other formats are supported only on the web for now. Users can also open documents in full-screen mode, highlight passages to ask Gemini questions, or request changes—such as turning a table into a graph.

Sources: The Verge, Tecnoblog

Read the original story ↗

Granola launched a dictation tool for Mac that works across apps. Users press the “Fn” key to speak, and the system cleans up and formats the text before inserting it into notes, prompts, emails, and messages. The company says the feature offers unlimited words for free and that versions for iOS and Windows will launch soon.

The release expands the competition with Wispr, which made the opposite move months earlier by adding meeting notes to its dictation app. According to TechCrunch, Granola and Wispr are valued at $1.5 billion and $2 billion, respectively. Fireflies, Speechify, and Monologue have also been combining dictation and meeting notes in multifeature apps.

Sources: TechCrunch

Read the original story ↗

The New York Times is testing an AI feature in the NYT Cooking app that suggests ingredient and equipment substitutions, scales recipes to any number of servings and adapts the preparation. Genevieve Ko, a senior editor at NYT Cooking, says the tool can, for example, advise when to add spinach instead of kale because it wilts faster. Customized versions are private and cannot be published or shared through the app.

The feature is rolling out gradually and is currently available to all Cooking subscribers on iOS; broader availability will come soon. Readers criticized AI’s environmental impact and its effects on the app experience. Ko said there is no option to disable it for now, although using it is optional. A New York Times spokesperson told The Verge that closing comments after four days is standard practice. The company had already used AI for measurement conversions and serving-size adjustments.

Sources: The Verge

Read the original story ↗

Death Clock, an app launched in 2024 to track health habits and estimate life expectancy, told the writer of a Wired article that they would die on August 10, 2060, at age 74. After the writer entered a mild family history of diabetes, the app changed the prediction to 73 and identified cardiovascular disease, a sleep disorder, or cancer as possible causes. The writer says that after a month, the experience felt as if it were draining their energy.

In February, the company announced a feature for ordering biomarker panels. The $69 subscription includes a basic blood panel, while nine additional tests cost $780 altogether; the writer also added a $29 stress test. Death Clock says it trained its AI on 1,200 longevity studies, actuarial data from life insurers, and life-expectancy tables to create personalized prevention plans.

Sources: Wired

Read the original story ↗

A MIT Technology Review Insights survey finds that, although every organization surveyed uses AI to some degree, only 16% are extracting transformative value from their deployments. Forty percent report moderate gains in specific tasks or teams, 21% see meaningful gains, and 23% report minimal or no returns. Most are still piloting use cases (38%) or using copilots and conversational assistants to support employees (35%); 12% have AI agents in production and 3% use AI to do most of the work in key processes.

Atlan co-founder Prukalpa Sankar says the gap between what models can do and their usefulness in companies stems from a lack of context. She describes a “context layer” as infrastructure between business systems and AI platforms that turns enterprise intelligence into governed, machine-usable context for models and agents. The material was produced by MIT Technology Review Insights, the publication’s custom-content arm, rather than its editorial staff.

Sources: MIT Technology Review

Read the original story ↗

A notification shown on an LG TV says Microsoft has decided to end Copilot support on all televisions during 2026. The message, revealed by Windows Central and reported by The Verge, says the feature is no longer available and may stop working shortly, but does not provide an exact date. It was not an official announcement from Microsoft, LG, or Samsung; the schedule depends on Microsoft’s timeline.

Copilot let users chat with the AI, ask questions about what was on screen, and receive movie and TV recommendations. Tecnoblog tested the feature on an LG TV in Brazil and confirmed that the integration was ending on models available in the country. The outlet also linked the decision to Microsoft’s shift toward positioning Copilot as a productivity tool; the integration no longer appears on Samsung’s Tizen systems.

Sources: The Verge, Tecnoblog

Read the original story ↗

IEEE Spectrum describes engineering projects as increasingly combining human judgment with machine capability. AI tools can search research and standards, generate design options, test them in simulations and compare results. Specialists still need to assess issues such as safety, buildability, the setting in which a design will be used and its impact on people, while the decision about whether to proceed remains human.

The article says AI does not eliminate the need for expertise in fields such as electrical, mechanical, civil and software engineering; it makes it easier to explore aspects outside each professional’s specialty. Systems thinking—understanding relationships, constraints, failure points and trade-offs—therefore becomes central. Engineers must also decide what to delegate to AI, what people should check and when to redefine the problem, since the tools can produce convincing answers based on weak assumptions or optimize a design for the wrong goal.

Sources: IEEE Spectrum

Read the original story ↗