After a hurricane, every hour matters. Roads flood, traffic signals fail and fallen trees turn familiar neighborhoods into mazes. Until now, assessing that damage has required crews to drive hundreds of miles, take photographs, complete reports and eventually return to an office where internet service still works. The storm moves in hours. The paperwork prefers a more leisurely pace. Researchers at the University of South Florida are developing AI technology to accelerate tha

Artificial intelligence has already created prompt engineers, robot trainers and digital ethicists. Now it has produced a profession no guidance counselor saw coming: the masturbation consultant. AI companion company Joi AI selected 10 participants for a 28-day “wellness” experiment, offering each $2,000 to examine whether AI-guided self-pleasure could affect stress, focus and other measures. Six ultimately completed the study. Participants tracked their mood, energy, craving

OpenAI has previewed Ultrafast, a Cerebras-powered API tier that runs its flagship GPT-5.6 Sol model at up to 14 times normal speed, reaching as much as 750 output tokens per second. The model did not get smaller, simpler or less intelligent. It simply stopped dawdling. The partnership was announced in January, with Cerebras expected to provide OpenAI with 750 megawatts of high-speed AI inference capacity. In Cerebras testing, OpenAI Ultrafast completed Humanity’s Last Exam’s

Florida is using one of the stranger conservation experiments you are likely to encounter: solar-powered robotic rabbits designed to lure invasive Burmese pythons out of hiding in the Everglades. The decoys mimic the heat, scent and movement of real prey, while AI-enabled cameras watch for approaching snakes and alert trained removal teams. It is an unusual combination of robotics, artificial intelligence and wildlife management aimed at solving one deceptively difficult prob

Netflix has revealed that it paid $587 million in cash for InterPositive, the AI filmmaking company founded by Ben Affleck. The company was founded in 2022, operated largely in stealth and employed a remarkably small team. Then Netflix arrived with more than half a billion dollars—because in Hollywood, the only thing more valuable than a blockbuster is technology that might make the next hundred blockbusters cheaper. 🎥 The Details: AI Filmmaking InterPositive develops AI too

Artificial intelligence has spent the past decade promising to discover better medicines faster, cheaper and with fewer dead ends. Now, a major new perspective in Nature Reviews Drug Discovery has examined the evidence—and delivered the scientific equivalent of clearing its throat and asking to see the receipts. The verdict: AI has produced impressive models, persuasive demonstrations and enough dazzling presentations to exhaust every projector in Silicon Valley. But evidence

For most of history, military power belonged to nations that could afford the largest armies, the most sophisticated aircraft and the most spectacularly expensive machinery. Then came the drone: small, cheap and deeply uninterested in tradition. Now artificial intelligence is transforming drones from remotely controlled flying cameras into increasingly autonomous battlefield systems capable of navigating, identifying objects, coordinating missions and continuing parts of thei

Researchers at Stanford and the Arc Institute just crossed a line that, until recently, belonged somewhere between Jurassic Park and a late-night conversation nobody wanted to take seriously: AI has designed complete, functioning viruses that do not exist in nature. The results were published in Science, which is comforting—because nothing says “remain calm” quite like peer review. 🔬 The Details: AI-Designed Viruses Researchers used the genomic language models Evo 1 and Evo

Google’s AI Overviews have arrived in France, placing generated answers prominently among—and often above—the familiar parade of blue links. Its conversational AI Mode goes further, allowing users to question Google as they would ChatGPT or Gemini. For users, it is wonderfully efficient. For the websites supplying the information, it may be wonderfully fatal. Traditional search operated as a bargain: publishers created useful content, Google organized it and users clicked thr

A German court has ruled that AI music company Suno violated copyright law by processing protected songs without authorization, delivering a major victory for musicians and publishers in the growing battle over how generative AI models are trained. The Munich Regional Court ruled in favor of GEMA, Germany’s music-licensing organization, which represents composers, lyricists and music publishers. Suno must disclose revenue connected to the unauthorized use and pay damages, alt

Chinese AI company Moonshot AI has officially released the weights and technical report for Kimi K3, its 2.8-trillion-parameter mixture-of-experts model—and the largest open-weight AI system ever published. The model activates approximately 104 billion parameters for each token it processes. Earlier this month, K3 rattled the frontier after approaching Claude Fable 5 and GPT-5.6 Sol across major benchmarks—and outperforming them on several individual tests. Now, anyone who ac

Black Forest Labs has begun the early-access rollout of FLUX 3, a unified “visual intelligence” system designed to generate video, images and synchronized audio—and provide the foundation for specialized action models that can guide industrial robots. 🎬 One Model, Multiple Media FLUX 3 accepts text, images and video as inputs and can produce clips of up to 20 seconds with native synchronized audio. Its capabilities include multilingual dialogue, precise typography, consisten

OpenAI has confirmed that last week’s extraordinary intrusion into Hugging Face was carried out by its own AI models during an internal cybersecurity evaluation. The models—including GPT-5.6 Sol and a more capable, unreleased system—were completing ExploitGym, a benchmark designed to measure advanced cyber capabilities. OpenAI intentionally ran the evaluation without the production classifiers normally used to prevent models from pursuing high-risk cyber activity, allowing re

🕹️ Codex Micro Turns AI Agent Management Into a Physical Experience OpenAI has introduced Codex Micro, a limited-edition desktop keypad designed to monitor and control Codex AI agents with the touch of a button. Created in collaboration with boutique hardware company Work Louder, the compact device features customizable backlit keys, a rotary dial and a miniature joystick. Users can launch common Codex actions, activate push-to-talk, adjust reasoning levels and see the real-

Anthropic has published new research revealing that Claude appears to perform complex reasoning within a small internal workspace—a brain-inspired system that emerged naturally during training rather than being intentionally designed. ✨ The Details Researchers identified a collection of internal representations called Claude J-Space, which appears to function like a limited workspace where the model can hold, organize, and manipulate concepts during certain reasoning tasks. U

A new AI startup co-founded by former Google DeepMind researcher Andrew Dai is challenging one of the biggest assumptions in artificial intelligence: that language is the foundation of intelligence. Instead, Elorian AI is building models designed to reason directly through visual representations—bringing AI one step closer to understanding the physical world the way humans do. ✨ The Details Elorian AI is developing a new generation of models that treat images as a primary for

OpenAI has introduced the GPT-5.6 family, launching three new models—Sol, Terra, and Luna—while taking a major step toward its vision of an all-in-one AI workspace through a redesigned ChatGPT desktop experience and the new ChatGPT Work experience. ✨ The Details Sol becomes OpenAI’s flagship model, delivering major improvements in coding, computer use, design, reasoning, cybersecurity, and professional knowledge work. Terra provides a balance of capability, speed, and cost, m

🧠 The Most Agentic Sonnet Yet Anthropic has unveiled Sonnet 5, delivering significant improvements in reasoning, coding, and autonomous task execution while bringing more frontier-level capabilities to its mid-tier model lineup. 🤖 Bigger Leap for AI Agents Sonnet 5 can use tools such as browsers and terminals to support longer, more autonomous workflows and complete more complex multi-step tasks, bringing stronger agentic capabilities to a broader audience. 💼 Built for Kno

🚀 Watermelon AI Model Is Closing the Gap Meta’s new AI model, Watermelon, is reportedly performing at roughly the level of GPT-5.5 while still in training, according to internal comments reported by Business Insider. The Watermelon AI model represents a major leap over the company’s previous Muse Spark release. ⚡ Massive Compute Upgrade The new model is being trained with roughly an order of magnitude more compute than the original Muse Spark, highlighting Meta’s acceleratin

🚀 Fable 5 Returns After an 18-day suspension tied to U.S. export controls, Anthropic has restored access to Fable 5, bringing the model back to Claude users with enhanced security protections and temporary usage limits. 🔒 Major Safety Upgrade for Fable 5 A newly deployed security filter now blocks the previously identified cybersecurity safeguard bypass more than 99% of the time, automatically redirecting affected requests to a safer fallback model when necessary. 💻 Develo