Today's Hallucination HQThe King Has Lost Its Majority (But Still Owns the Building)
ChatGPT's market share has dipped below 50% for the first time — a historic milestone that sounds more alarming than it is. The chatbot still commands 1.1 billion monthly users, which is more than the entire population of the Americas. Gemini trails at 662 million, Claude at 245 million. OpenAI remains comfortably dominant; it's just that "comfortably dominant" now means slightly less than everything.
Source: TechCrunch
Elon Musk Buys the Coding Tool. Just a Small One. Sixty Billion Dollars.
Fresh from its IPO, SpaceX has agreed to acquire Cursor — the AI coding assistant — for $60 billion, which is either an extraordinary vote of confidence in developer tools or a receipt that will age poorly. The move is designed to help Musk's ever-expanding empire of rockets, chatbots, and social media arguments compete with Anthropic and OpenAI in the enterprise market. Cursor was not entirely surprised. Neither, frankly, were we.
Source: The Verge
Picked a Fight With the Government. Sales Are Up.
Anthropic's ongoing dispute with the Trump administration appears to be doing wonders for its bottom line. Spending data from Ramp suggests enterprise adoption of Claude is growing robustly — which implies that some businesses view a company willing to argue with Washington as a reassuring sign of backbone rather than a liability. Getting into a public row with the government and emerging more popular is a genuinely rare trick. Well played.
Source: TechCrunch
Microsoft's AI Assistant Briefly Moonlighted as a Hacker's Assistant
A now-patched vulnerability in Microsoft Copilot — dubbed SearchLeak — allowed attackers to steal two-factor authentication codes from users, which rather undermines the point of two-factor authentication. The exploit is the latest in a dependable series demonstrating that the industry's approach to securing large language models follows a recognisable pattern: ship, get hacked, patch, repeat. Security researchers described it as critical. Microsoft described it as resolved. Both are correct.
Source: Ars Technica
Your AI Travel Agent Has… Interesting Taste in Itineraries
A new benchmark paper from arXiv tests whether AI agents — not just chatbots answering questions, but systems actively booking travel and planning logistics on your behalf — will schedule activities involving animal suffering if you don't explicitly tell them not to. Spoiler: several will book you a bullfight without flinching. The research matters because agentic AI (AI that acts, rather than just advises) is arriving fast, and "didn't know the user cared about ethics" is not a policy.
Source: ArXiv AI
The AI industry: moving fast, breaking things, occasionally booking you a bullfight.
|