Today's Hallucination HQAnthropic Releases Claude Opus 5, Politely Suggests It Might Be Quite GoodAnthropic has launched Claude Opus 5, its most capable model yet — one that the company says performs at or above human level on a range of demanding benchmarks. It scores particularly well on complex reasoning, coding, and research tasks. The system card notes meaningful improvements in honesty and instruction-following, which is either reassuring or a quiet admission about previous versions. Available via API and Claude.ai, with pricing that reflects Anthropic's confidence in the thing.
Source: Anthropic
Google's Traffic Tap Runs Dry, Publishers Discover They Were the PipesFor decades, Google and the web had a gentleman's agreement: Google indexed your content, you got visitors, everyone pretended it was a partnership. AI Overviews — Google's feature that answers questions directly on the search results page — has quietly ended that arrangement. Publishers are now reporting dramatic drops in referral traffic, a phenomenon being called "Google Zero." The web built itself around Google's appetite. Turns out, Google just got a meal delivery service.
Source: The Verge
Midjourney Buys Astrology App, Stars Allegedly FavourableMidjourney, the AI image generation company, has acquired Co-Star, the popular astrology app with tens of millions of users. Midjourney's stated ambition is to build tools that model the "human experience" — and apparently the human experience includes checking whether Mercury is in retrograde before replying to emails. Co-Star brings a substantial, notably young user base and rich personal data. Whether AI-generated art and birth chart readings belong in the same product is a question the market will answer, presumably during a full moon.
Source: TechCrunch
Meta AI Learns to Read Your Calendar, Promises Not to Judge Your SchedulingMeta is upgrading its AI assistant with productivity features including calendar access, task management, and document handling — bringing it broadly in line with what ChatGPT, Gemini, and Claude already offer. The update is being positioned as a push into "agentic" territory, meaning the AI can take actions on your behalf rather than just answering questions. Meta has roughly three billion potential users for this across its apps, which is either a significant advantage or a tremendous amount of pressure on a chatbot.
Source: The Verge
Washington Targets Chinese Humanoid Robots, Because Chips Alone Were Too SimpleThe US government is expanding its technology restrictions on China to include humanoid robots and other autonomous systems, citing national security and data privacy concerns. This follows years of controls on semiconductor exports and AI model access. The concern is straightforward: robots operating in American environments could collect sensitive data and send it home. It is, when you think about it, a remarkable moment — one in which the phrase "and another thing" is doing a great deal of geopolitical heavy lifting.
Source: Digitimes
As ever, the AI industry moves fast — mostly so you don't have time to ask questions.
|