Latest AI News

A study shows that nearly 7,000 global data centers are built outside the ASHRAE‑recommended temperature range of 18–27 °C, forcing continuous cooling and lowering energy efficiency. Site selection is driven mainly by power availability and cost rather than climate suitability, leading to many facilities in hot regions such as Singapore, Nigeria, and the UAE where temperatures often exceed 27 °C. Most centers use air cooling; liquid cooling is only emerging for high‑density racks. This mismatch increases strain on local power grids and resources. With AI workloads expected to double energy consumption by 2030, addressing the climate misalignment is essential for sustainable data center operations, as continued reliance on inefficient cooling could threaten worldwide viability.

The author posted an unexpected divorce announcement on Instagram with a photo and text. Days later she discovered that an AI had appended additional content to her post after it was published. The added text mimicked her writing style, incorporated SEO keywords, and reframed the story as a self‑help narrative. She objects to this unauthorized use of her voice, citing violation of authorship and potential impersonation. The incident highlights Meta’s lack of transparency about AI‑generated content and no safeguard against such additions. Emotionally, the AI altered the original context, removing details of betrayal, financial insecurity, and vulnerability, and replaced them with generic therapy language that trivializes her experience. The author argues this reflects biased internet infrastructure that dismisses women’s pain and erases marginalized narratives, underscoring the need to address biases in AI systems.

Intel released Generative AI Examples v1.5, an open‑source showcase built on the Open Platform for Enterprise AI (OPEA). The update adds new use cases such as automating web tasks and summarizing legal documents, introduces a Polylingua translation service, and provides compatibility with the OpenAI API. Technically, it supports Intel Arc GPUs through LLM‑Scaler‑vLLM and offers a drag‑and‑drop interface in GenAI Studio for fine‑tuning applications. Hardware validation uses 3rd‑generation Xeon Scalable processors (Ice Lake, 2021) and Gaudi Gen2 AI accelerators rather than newer CPUs or GPUs. The project is hosted on GitHub.

OpenAI’s Sora 2 video generator is being misused to produce disturbing, often sexually suggestive content featuring AI‑generated children. Examples include fake toy commercials with suggestive designs and videos parodying real events that exploit child imagery. While the material is not always explicit pornographic, it raises concerns about exploitation and predatory behavior. OpenAI has built safeguards against deepfakes of minors and bans CSAM, yet creators are circumventing these measures; some accounts have been banned but problematic videos still surface on platforms such as TikTok. The challenge lies in distinguishing harmless depictions from exploitative content, requiring more sophisticated moderation, keyword limits, and better training for moderators. UK and US lawmakers are considering legislation to prohibit AI‑generated child sexual abuse material, underscoring the regulatory difficulties of protecting vulnerable individuals online.

Waymo’s robotaxi service in San Francisco resumed after a Saturday power outage caused traffic light failures and gridlock. The blackout was triggered by a PG&E substation fire that disabled Waymo’s automated driving systems, forcing vehicles to stop at intersections. Although the Waymo Driver can treat non‑functional signals as four‑way stops, the widespread outage required extended pauses for confirmation, contributing to congestion. Service was suspended during the peak of the outage and Waymo coordinated with city officials before restarting operations. The incident illustrates how large‑scale infrastructure failures challenge autonomous vehicle operation compared with human‑controlled ride‑hailing services like Tesla’s.

Samsung will launch its Bespoke AI Refrigerator Family Hub at CES 2026 with Google Gemini AI integrated. The refrigerator uses an internal camera to detect food items and quantities, providing meal‑planning tips, shopping lists, and spoilage alerts. Voice control enables hands‑free operation, while the AI adjusts cooling for energy efficiency based on usage patterns. A similar Gemini‑powered Wine Cellar is also planned. This move extends Gemini’s capabilities into home appliances, using vision and voice to reduce food waste and lower energy consumption.

In 2025, 20 % of Google’s new AI software engineers were “boomerang” hires who had left the company for a period before returning. This trend has accelerated since 2022 and mirrors a wider industry practice of poaching talent from rivals such as Microsoft and Apple. The surge is linked to Google’s intensified focus on AI, highlighted by the launch of Gemini 3 and its growing user base of 650 million monthly users. Although still behind ChatGPT in weekly active users, Google shares rose 62 % this year, reflecting investor confidence. Meanwhile, Apple has also recruited seasoned AI professionals, including a recent former Google engineer who became its VP of AI. The competitive landscape in the AI sector is driving these high‑profile hiring moves across major tech firms.

Veeam CEO Anand Eswaran says that across all industries, data security and resilience are essential for successful AI. He warns of a growing threat landscape and the fragmentation of tools used for data management in the age of AI. Veeam’s unified platform is designed to give end‑to‑end protection—data security controls, privacy governance, and resilience from creation through AI pipelines. Eswaran argues that trustworthy AI depends on these safeguards and that projects fail when they lack them. He advocates a cautious, pragmatic approach to AI adoption, prioritizing precise use cases, and stresses that Veeam’s platform removes the data‑foundation barriers that prevent organizations from fully exploiting AI.

The article discusses how to scale Large Language Models (LLMs) for large codebases by tackling verification challenges. It stresses the need for “oversight” through programmatic checks, safety mechanisms that protect language abstractions, and automated QA processes to reduce bottlenecks. By automating feedback loops, LLMs can iteratively refine their outputs. The piece also argues that building modular systems and reusing existing components—such as code libraries—are preferable to developing from scratch. Overall, the goal is to enhance LLM reliability and ease of adoption by programmers through stronger quality control and lower friction in development workflows.

A recent study discovered “H‑Neurons,” a small subset of neurons in large language models that consistently predict when the model will hallucinate. These neurons are causally linked to over‑compliance behaviors, leading to factually incorrect outputs. They emerge during pre‑training rather than being solely data artifacts, indicating they are an intrinsic part of the architecture. Mapping and targeting these H‑Neurons could enable more reliable LLMs by addressing the neural mechanisms that generate hallucinations.

Instacart has halted its AI‑powered pricing experiments after finding they produced inconsistent prices for the same products at a given store and time. The company will no longer allow retailers to use its Eversight AI tool for price testing, citing that the trials were not driven by supply/demand or customer data. Lawmakers and the FTC had scrutinized the practice, and Instacart recently settled with the agency for $60 million over deceptive advertising claims. The firm acknowledged the tests “missed the mark” for some shoppers and will discontinue the program.

iRobot filed for Chapter 11 bankruptcy after years of innovation in robotic cleaning. The company’s flagship Roomba could not reliably perform in real homes—pet hair, clutter, and varied flooring thwarted its sensors and navigation. While advances in AI and robotics have been notable, the technology still struggles with unpredictable domestic environments. The article argues that consumer appliances are poorly suited for seamless smart‑home integration, creating a large gap between advertised capabilities and actual user experience. Investors now focus on more distant robotic projects, such as humanoid robots, rather than refining Roomba’s practicality. The Roomba’s collapse highlights the limits of current technology when confronted with everyday messiness.

AI demand has driven hyperscale data‑center operators to boost infrastructure spending and capacity dramatically. Over the past three years, capacity additions rose 170 %, while capex climbed nearly 180 % to $142 billion in Q3 of this year. Global datacenter capacity is now almost triple that of 2018, with facilities reaching hundreds of megawatts. The United States holds 55 % of hyperscale capacity. A pipeline of 770 new facilities indicates continued expansion, but if AI investment slows there could be an oversupply risk and potential capacity contraction.

OpenAI’s reports of child sexual‑abuse material (CSAM) to the National Center for Missing & Exploited Children (NCMEC) rose sharply in early 2025, reaching 80 times the volume seen in the same period in 2024. The spike aligns with new product features such as image uploads and increased overall user activity. OpenAI reports all CSAM—including uploaded images and text requests—and has expanded its review capacity while adding safety tools: parental controls, self‑harm detection, threat monitoring, and a Teen Safety Blueprint that details ongoing CSAM detection improvements. The surge mirrors a nationwide rise in generative‑AI‑related CSAM reports. These developments come amid heightened scrutiny from state attorneys general, congressional hearings, and FTC investigations focused on AI safety for children.

The Indie Game Awards revoked the Game of the Year and Debut Game titles from Clair Obscur: Expedition 33 because the developers used generative AI in production, a practice prohibited by the awards rules. Although the team claimed the AI was only for placeholder textures that were later removed, some AI‑generated assets remained in the final game. The organization cited a breach of its regulations despite prior statements from the developer denying any AI use. Consequently, Blue Prince received Game of the Year and Sorry We’re Closed received Debut Game. This reversal contrasts with Clair Obscur’s success at The Game Awards and its commercial performance, having sold more than five million copies.

Visa plans to launch AI shopping and payment agents on its network by 2026, allowing automated purchases for consumers and businesses. Current pilots run in controlled settings, while the company has introduced a “Trusted Agent Protocol” so merchants can verify legitimate AI agents. Akamai supplies additional bot‑detection and identity‑theft safeguards. Although Visa says the technical infrastructure is ready, the rollout raises concerns about consumer awareness of risks when AI handles financial transactions.

Samsung has added Google Gemini AI to its refrigerators, enabling features such as recipe suggestions, shopping list creation, and spoilage monitoring. The appliance tracks door openings, food consumption patterns, and uses computer vision to identify contents, generating scores for energy efficiency and nutrition that could influence insurance rates. While the company touts improved food management and convenience, consumers express concerns over data privacy, potential malfunctions (e.g., compressor shutdowns), and added cost. Many view smart appliances as distractions from core functionality and prefer simpler models. The success of this integration will depend on delivering tangible value without compromising user privacy.

Retro’s new app Splat uses generative AI to turn photos into printable coloring pages for kids. Users upload a photo or choose from pre‑made categories such as animals and cartoons, then the AI converts it into a line drawing with styles like anime or comic book. The app is available on iOS and Android and offers weekly ($4.99) and annual ($49.99) subscriptions to generate pages. A parental authentication step protects settings from child access. Splat joins other AI‑driven creative tools, such as sticker generators and robotic pets that learn personalities, aimed at encouraging children’s creativity.

In 2026 Uber and Lyft will start testing Baidu’s Apollo Go robotaxis in London, joining Waymo and Wayve as city operators. The tests involve Baidu’s electric RT6 SUVs equipped with its autonomous driving system. Lyft plans to expand the fleet to hundreds of units once regulatory approval is obtained; Uber expects to begin trials in the first half of 2026 under its existing partnership with Baidu. This collaboration highlights ride‑sharing firms’ strategy to partner with autonomous vehicle developers to grow their robotaxi services worldwide.

OpenAI’s ChatGPT launched in November 2022 and now powers over 800 million weekly active users, with a free tier and paid Plus plan that includes advanced voice, image generation (GPT‑4o) and web browsing. Recent model releases—GPT‑5.1, GPT‑4.1, o3, and o4‑mini—improve reasoning, coding, and cost efficiency; GPT‑5‑Codex and GPT‑5‑Codex‑C enhance developer workflows. ChatGPT’s enterprise focus includes business agents, cloud integrations, meeting recording, research connectors, and the Apps SDK for custom apps. Partnerships span Apple (Apple Intelligence), Disney (exclusive character use in Sora videos), Walmart (product recommendations), Reliance Industries (API distribution in India), and Microsoft/Bing integration. OpenAI expands globally with new markets in Asia, a $6.4 billion hardware acquisition, and data residency programs. Safety upgrades target teen interactions, mental‑health risks, and content filtering; legal challenges involve copyright lawsuits over music lyrics and image generation. Revenue is projected at $12.7 billion this year, driven by the mobile app’s $2 billion earnings and growing business subscriptions.

Apple’s iPhone quietly outperforms rivals like Google and Samsung in everyday AI with “Siri Suggestions.” By analyzing user habits and location data, the feature anticipates needs—suggesting relevant apps (e.g., train schedules) or automatically adding event details to the calendar. This practical machine learning streamlines app access and automates tasks, saving time for users. The article highlights how this subtle integration demonstrates Apple’s focus on enhancing daily smartphone use rather than relying on headline‑grabbing AI gimmicks, indicating a positive direction for future iPhone AI development.