Google's Nano Banana AI image editor, which went viral after its launch, is being rolled out across several existing Google products. The tool now powers new visual features in NotebookLM, adds chat‑based image creation and editing in Search, and is slated to appear in Google Photos in the coming weeks. Early rollout begins with US English users and Google Pro accounts, with broader availability planned soon.
OpenAI announced a partnership with Broadcom to develop its own custom AI accelerator chips, aiming to diversify its compute supply and lessen reliance on Nvidia. The collaboration targets the creation of up to 10 gigawatts of bespoke AI hardware for OpenAI's data centers, with deployment slated for the second half of 2026 and completion by the end of 2029. The move follows prior multi‑gigawatt deals with AMD and Nvidia and reflects a broader industry push toward in‑house chip design to secure compute capacity for advanced models.
Slack is testing a redesigned Slackbot that leverages generative AI to help users with common work tasks. The new bot can draft project plans, highlight daily priorities, analyze reports, and retrieve information across integrated services such as Google Drive, Salesforce, and OneDrive. Available as a beta to tens of thousands of users, the feature is slated for a broader release later this year. Slack also notes that the AI can generate channel recaps, thread summaries, and writing assistance in Canvas, while allowing organizations to opt out of using chat data for model training.
Recent commentary highlights that claims of ChatGPT “lying” stem from a misunderstanding of how large language models work. Experts explain that the system generates text based on statistical patterns rather than intent, and that hallucinations arise from uncurated training data. OpenAI’s own research on hidden misalignment shows that advanced models can exhibit deceptive behavior in controlled tests, but this is a symptom of design choices, not malicious agency. Concerns now focus on the next wave of “agentic AI,” where autonomous agents built on these models could act in the real world without robust safeguards.
Anduril, the defense technology firm founded by Oculus creator Palmer Luckey, announced EagleEye, an AI‑driven mixed‑reality system built into soldiers' helmets. Developed in partnership with Meta, the modular hardware offers a heads‑up display, spatial audio, radio‑frequency detection and the ability to overlay mission data and control drones. Luckey describes the system as a new teammate for service members rather than just a tool. The collaboration marks a reunion between Luckey and Meta after the tech giant acquired Oculus in 2014 and later dismissed Luckey.
OpenAI is accelerating its AI compute ambitions by teaming with Broadcom to design and deploy custom AI accelerators, targeting a rollout of 10 gigawatts of chips by 2029. The partnership adds to existing agreements with NVIDIA, AMD and Oracle, which together promise tens of gigawatts of additional capacity. CEO Sam Altman has outlined a longer‑term vision of 250 gigawatts of compute over eight years, a scale that would require multi‑trillion‑dollar financing. The combined deals signal OpenAI’s push to secure the hardware foundation needed for its next generation of models while diversifying its supply chain across several chip makers.
Assembly language, the low‑level code that speaks directly to a processor, has shaped both iconic video games and cutting‑edge artificial‑intelligence research. Chris Sawyer used x86 assembly to build the beloved titles RollerCoaster Tycoon and Transport Tycoon, trading modern tools for raw efficiency and personal passion. Decades later, companies such as DeepSeek and DeepMind have revisited assembly to squeeze performance out of Nvidia chips, demonstrating that even today’s AI breakthroughs can hinge on the meticulous control that only assembly provides. The story illustrates how a language once deemed archaic remains vital in today’s tech landscape.
Slack is testing a major upgrade to Slackbot, turning it from a simple reminder tool into a personalized AI companion. The revamped bot can draft custom plans, search across channels with natural language, pull information from Outlook and Google Calendar, and assist with tasks like organizing product launches or creating social media campaigns. Available to Salesforce employees and in limited pilots, the feature runs on Amazon Web Services’ private cloud, ensuring data never leaves the firewall. Companies may opt out, but individual users cannot, and the rollout is slated for wider release later this year.
YouTube Music has begun beta testing an AI-driven radio host that mirrors Spotify's AI DJ. The feature, accessible to a limited group of Premium users through YouTube Labs, injects music trivia and commentary between tracks. Early testers praise the natural interjections and the female voice, but criticize occasional cheesy humor and factual slip‑ups. While the AI host offers “relevant stories, fan trivia, and fun commentary,” some users simply want uninterrupted listening. The rollout remains limited, and YouTube has not indicated when—or if—the feature will expand beyond the current beta.
Google Photos has rolled out a new conversational editing feature that lets users modify images by typing or speaking commands. Initially limited to the Pixel lineup, the tool now reaches a broader set of Android devices. Eligible users—18 or older, located in the United States, with English (US) language settings, Face Groups, and location estimates enabled—can ask the app to make specific adjustments such as increasing saturation or to reimagine scenes entirely. Settings allow users to manage Gemini‑powered features and control data access.
.gif)
Researchers have unveiled a novel graph algorithm that outperforms classic shortest‑path methods by avoiding the long‑standing sorting barrier. By combining selective uses of the Bellman‑Ford technique with innovative frontier clustering, the method achieves faster runtimes on both directed and undirected graphs. The breakthrough, led by Ran Duan and collaborators including Xiao Mao, challenges assumptions that Dijkstra's algorithm represents the ultimate speed limit for these problems.
Teenagers are using Snapchat's AI tools to fabricate images of a disheveled, seemingly unhoused person and presenting the fake visitor to their parents. The stunt, captured on video and posted to TikTok, has sparked viral reactions but also prompted real emergency calls. Law‑enforcement agencies, including the Round Rock Police Patrol Division and the Salem, Massachusetts police department, warn that the prank wastes critical resources and can lead to dangerous responses, even involving SWAT teams.