AI that fits your day.

Google announced that Gemini, Gemini Notebook, Gemini in Classroom, and Search will be disabled by default for K–12 student accounts until administrators enable them individually. The company says this will give schools “more choice, not less,” but it will not offer a separate switch for AI Overviews and AI Mode, which are integrated into search. Schools can therefore keep search while disabling the Gemini apps; to avoid any AI feature in search, they will also have to block the search engine for students starting in 2027, according to Xataka.

Educators and parents criticize the policy because the summaries may reduce students’ source comparison and independent effort, while also producing errors. Google says it is updating search with Guided Learning, which guides students through problems step by step instead of simply providing answers. According to g1, 150 million students and teachers worldwide use Google Classroom.

Sources: Wired, g1, Ars Technica, Xataka

Read the original story ↗

Nothing CEO Carl Pei said on X that the company will ship its first “AI-native” product in December. He described the device as a productivity tool with Nothing’s personality, without a new screen and one that “couldn’t have been an app,” but gave no details about its form or an exact release date.

The announcement comes a little over a year after Pei outlined plans for “AI-native” devices and an AI operating system. According to TechCrunch, Nothing has also said it plans to release AI-linked smart glasses and earbuds with AI integrations. The Verge cites earlier rumors about two versions of the smart glasses and notes that the new product joins other screen-free AI devices already announced in the market.

Sources: The Verge, TechCrunch

Read the original story ↗

ChatGPT can now create GIFs from a prompt or a photo uploaded to the chatbot. The update lets users adjust animation speed, edit individual frames, and share the result — capabilities that were previously unavailable in a single native step in ChatGPT. Images 2.6 is also rolling out to ChatGPT and Codex users on the web, desktop, and mobile apps.

OpenAI says the model speeds up image generation and that users generate more than 4 billion images per week across ChatGPT and its API, up from the 3 billion per week the company reported in September. The company also launched Images 2.6 Flare and Sunburst in its API for faster, higher-quality image generation.

Sources: The Verge, OpenAI

Read the original story ↗

The Verge’s Victoria Song writes that the Amazon Bee Frontier, an AI wearable that records conversations and transcribes and summarizes them, has begun affecting her trust in her own memories. After asking for consent to use it during therapy sessions, Song says her therapist eventually recommended that she throw the device out the window.

Song says the device has improved at distinguishing her speech from television dialogue and can now access her calendar and photos app, as well as suggest tasks and purchases. Still, she writes that its summaries often incorrectly assign statements or opinions to her — including a supposed recurring argument with someone she has never met and six cats that are actually nicknames for her two animals. For the columnist, the errors can be funny but become disturbing when they involve serious conversations.

Sources: The Verge

Read the original story ↗

MIT Technology Review Insights says adopting AI agents requires tracking data as it moves among systems, files, and agents. In a scenario described by the publication, one agent combines restricted pricing, internal communications, and archived quotes; another uses the result to prepare a proposal, without human review, that includes confidential and unapproved information. Without data lineage, the company cannot identify where the problematic content entered the workflow or how it spread.

The report argues that security approaches focused only on applications, systems, networks, or endpoints are no longer sufficient for workflows in which data is continuously ingested, copied, summarized, transformed, reused, and redistributed. The material was produced by MIT Technology Review’s custom content arm, not its editorial staff, with research and writing conducted by humans.

Sources: MIT Technology Review

Read the original story ↗

Photon Matrix Lab, based in Changzhou in China’s Jiangsu province, has developed the Laser Mosquito Air Defense System, a roughly brick-sized device that combines lidar, cameras and artificial intelligence to locate small flying objects. According to Xataka, it emits about 100,000 low-power laser pulses per second, calculates a target’s trajectory and fires a higher-power laser to burn the mosquito’s wings; each shot lasts about one-hundredth of a second.

The company offers a version with an invisible infrared laser, mainly intended for indoor use, for US$988, and another with a visible blue laser for outdoor use at US$1,088. The concept dates back to Photonic Fence, a project presented by Nathan Myhrvold and engineers in the United States in 2009 and publicly demonstrated in 2010, but it never reached the mass consumer market.

Sources: Xataka

Read the original story ↗

In a g1 test, Alexa+ began recognizing the reporter’s name correctly and showed more elaborate conversations, better handling of unpredictable questions and the ability to search the internet. When asked about mosquitoes in Poços de Caldas, Minas Gerais, the assistant cited a municipal mosquito-eradication program and recommended bringing repellent—information the reporter checked. On an Echo Show 8 with an 8.6-inch display, it also showed weather forecasts, smart-home information, photos and alerts from a camera-equipped doorbell; to play the “O Assunto” podcast, however, the reporter had to specify the g1 program.

The experience had limitations. Alexa+ repeatedly displayed unsolicited recipe suggestions, which, according to g1, created the impression of a storefront for purchases from Amazon’s store, and restricted the number of photos that could be used in its digital photo-frame function. The reporter also chose not to connect email to the device because of privacy concerns, leaving announced features such as summarizing documents and creating travel itineraries untested.

Sources: g1

Read the original story ↗

Nexus Mods announced that authors of new mods must disclose whether AI was used in their work. Mods tagged as “AI-assisted” must also include a public explanation of how and where the technology was used, while the labels will become more prominent on mod pages. Site owner Victor “Foledinho” Folmann said the platform is “pro-human creativity,” but will not introduce an outright ban because AI may be used across several stages of creation, from script autocompletion to generating textures, voices, and other assets.

Nexus Mods says it will not use mods to train AI or sell or license them for that purpose. The platform is also preparing controls to hide AI-generated content on game pages, the homepage, and eventually in search. Authors will have 60 days to update existing mods; the grace period ends on January 2, when the rules will begin to be enforced.

Sources: The Verge

Read the original story ↗

Apple announced that developers can integrate Siri AI’s new features into “all apps” as part of the iOS 27.3 Developer Preview, which began rolling out on Tuesday. The assistant can use a user’s personal context, understand natural-language requests, and use on-screen awareness to complete actions in third-party apps. The update also includes “app skills,” which let developers combine several tasks into a single action.

According to Apple’s documentation, users will not need to mention an app’s name when it is in the foreground, has an active Live Activity, or is the subject of a conversation with Siri AI. Apple is also bringing the expanded capabilities to its own apps, including Calendar, Files, Maps, Messages, Photos, Safari, and Shortcuts; the company says users can, for example, find a coffee shop they visited the previous week or archive files that have not been opened for more than a year.

Sources: The Verge

Read the original story ↗

g1 found that a photo shared as if it showed the Learjet 60 moments before it crashed in a residential area of Florida is false. Posted on Instagram, Facebook and X, the image shows a small jet about to hit a road filled with vehicles. OpenAI’s AI detector said it was generated with the company’s tools and identified the SynthID watermark, although it did not find reliable C2PA credentials.

The image resembles a fragment of the original crash video, recorded by cameras in passing cars, but the footage does not show the aircraft clearly. According to g1, the AI tool may have been used to “enhance” that material. The Learjet 60 had departed from Austin, Texas, for the Bahamas; all nine people aboard died.

Sources: g1

Read the original story ↗

Netflix has started rolling out its experimental AI Discover tab in the U.S., initially for people who use the service in English on supported TVs. The first feature offers recommendations organized around themes such as “Something Light,” “Scare Me,” “Family Movie Night,” and “Quick Watches,” based on mood, viewing preferences, available time, and what is new or trending. Netflix plans to bring the tab to more users in the coming weeks.

After choosing a theme, users see matching titles and can refine the search or restart with “Something Different.” The “Tune Your Recommendations” option lets them save recommendations or exclude titles so they are not suggested again. Netflix says the feature uses a new intelligence layer that responds to what users tell it in real time and keeps learning. The company describes the tab as a testing environment for deciding which features to keep or improve before bringing them to the main experience.

Sources: The Verge, TechCrunch

Read the original story ↗

NVIDIA says AI is changing engineers’ work in semiconductor fabs by speeding up simulations, predicting how materials and processes will behave, and testing changes in virtual environments before physical experiments. The company says this approach lets engineers evaluate more alternatives, identify problems earlier and spend less time waiting for results, while keeping decision-making with the engineers.

NVIDIA cites partner examples: COMSOL’s software reportedly achieved up to a 10x speedup for multiphysics simulations with the cuDSS library; Applied Materials’ Ginestra platform reportedly ran up to 10x faster, while its ACE+ platform reached up to 35x in topography simulations. Integrating cuLitho into Siemens’ Calibre reportedly delivered up to 17x acceleration on lithography workloads. Lam Research also demonstrated, according to NVIDIA, plasma-behavior predictions in milliseconds, compared with days for a full simulation.

Sources: NVIDIA

Read the original story ↗

Kenyan startup Fastagger, founded by Mutembei Kariuki, Jude Mwenda, and Stephanie Njerenga, developed Auni to work offline and on low-power phones. Launched in 2024 and integrated into the M-Pesa mobile payments app, the assistant analyzes transaction data and recommends marketing actions for small businesses such as salons, restaurants, and small manufacturers. About 6,000 businesses have signed up, according to IEEE Spectrum.

The company adapted an optical character recognition model to extract data from financial statements generated by mobile-money apps, whose PDFs can have poor quality and unfamiliar formats. Fastagger also moved parts of the processing onto the device, adapting the system to unreliable connections and the basic phones common among its users. Auni can identify busy periods and loyal customers, and help business owners send personalized messages to customers who have stopped returning.

Sources: IEEE Spectrum

Read the original story ↗

Apple has started making four Audio Intelligence features available in developer betas for the Apple Watch Series 12: important-sound detection, song identification, Live Rewind, and Siri Recaps. Live Rewind displays the previous 15 seconds as text after a double-press of the Digital Crown, while Siri Recaps uses “ambient listening” to create titles and key points from conversations. Apple says raw audio is processed in dedicated hardware, is not saved as a file, and cannot be accessed by the operating system, apps, or Apple. Saved snippets and summaries sync end-to-end encrypted when users have iCloud with two-factor authentication and a device passcode.

In a TechCrunch test, Siri Recaps captured useful details about prices, timing, and policies, but also mixed in household topics and recorded conversations the tester considered unimportant, underscoring privacy concerns about devices that are always listening.

Sources: TechCrunch, The Verge

Read the original story ↗

Google launched Dreambeans in India, its first market outside the U.S., for Android and iOS, with support for 13 local languages plus English. The app uses data from connected services including Gmail, Photos, Calendar, Search, YouTube and Gemini to generate personalized illustrated stories without requiring a prompt; some can be animated with the Veo model. Google says it chose India because of user interest and the overlap in usage across Gemini, Gmail and Google Photos.

Users can choose which services to connect, delete their data and ask Dreambeans to avoid certain subjects, Gozde Oznur, a Google product manager, said. Stories can be generated in only one language at a time. Sensor Tower estimates shared with TechCrunch put Dreambeans at nearly 55,000 U.S. downloads since June, but about 5,500 in October, an 84% monthly decline.

Sources: TechCrunch

Read the original story ↗

Shigeru Miyamoto, creator of Super Mario and The Legend of Zelda, said generative artificial intelligence is unlikely to completely replace people when genuine talent is involved. In remarks reported by VGC after he received Japan’s Order of Culture, he compared the technology with the shift from typewriters to word processors and said automation could create more room to reflect on what truly matters in a creation. Miyamoto also argued that the creator’s intent is essential and warned that work based only on copying and pasting rarely stands the test of time.

Xataka places his view in the context of growing AI use in game development and cites the negative reception to Codex Mortis, a demo made entirely with AI. In Miyamoto’s view, the tool can help create a game, but it does not replace the ideas and creativity that give it value.

Sources: Xataka

Read the original story ↗

According to The Information, cited by Ars Technica, Apple is reportedly developing the J595 home robot, which could be called HomeView Ultra. The device would be able to move around the house, support FaceTime calls and switch on lights when it detects someone entering a room. The company is reportedly targeting a 2027 launch at $1,000 or more, but development delays could push the debut to 2028.

The report says new Apple CEO John Ternus is apparently pushing the company’s robotics projects. The first prototypes appeared in 2018, and Tim Cook reportedly approved continuing the project. Engineers are said to be working to expand the robot’s capabilities to justify its price, while its reported features still appear limited compared with existing smart-home devices and robot vacuums.

Sources: Ars Technica

Read the original story ↗

OpenAI said in an email to users that it will begin testing an experience that combines ChatGPT’s Chat and Work tabs before expanding it. The proposal is intended to let users move from exploring an idea to carrying out tasks—including analyzing files, creating presentations or websites, and taking actions on the web—in one place. Codex will remain a separate experience in the ChatGPT desktop app.

According to Xataka, the allowance change will begin in December 2026 for the Plus, Pro, and Business plans. ChatGPT and Codex will draw from the same allocation; OpenAI says this will not affect most users, but advanced models and complex operations may consume more of the allowance and cause some users to wait longer for resets. The company has not yet detailed how much each task will consume.

Sources: TechCrunch, Xataka

Read the original story ↗

Wired says tools such as Claude, ChatGPT, and Gemini can make common searches easier, including widely known facts, as well as complex queries that require gathering information from multiple sources. AI can also interpret the question and present a synthesized answer instead of only a list of links. The result still depends on the quality and interpretation of the sources it finds.

According to Wired’s analysis, obscure queries, highly specific requests, and searches requiring precision—such as a quotation, a report statistic, or a device specification—may be better served by a traditional search engine. The publication recommends checking sources and carrying out an additional verification, because chatbots can answer confidently even when they fail to find reliable material or combine information incorrectly.

Sources: Wired

Read the original story ↗

In a report published by Xataka, a user tested Instinct to deregister an old motorcycle they no longer used. The agent explained that the vehicle could be taken to a scrapyard, found three nearby options offering home collection, provided their costs and helped choose one that deducted the scrap value from the towing service.

After the user’s approval, Instinct drafted and sent an email to the scrapyard and followed up on costs, terms and dates, requesting approval for each next step. The process took less than ten minutes, according to the report. The user notes that the service requires a phone number and recommends limiting the data and permissions given to AI agents, especially when sensitive information is involved.

Sources: Xataka

Read the original story ↗