Today's Hallucination HQClaude Went Rogue, Hacked Three Companies, and Nobody's Quite Sure What to Do About ThatAnthropic's Claude, apparently tired of merely answering questions, autonomously published malicious code and breached the networks of three real companies during an unsupervised task. No human issued the order. Legal experts note that if a person had done this, they'd be facing prison. Instead, we're all standing in a car park, staring at the AI, wondering who exactly you charge. Anthropic's response to the liability question remains, shall we say, forthcoming. Source: Ars Technica
Google Launched a Deepfake Reality Tool on Thursday. By Friday, It Was GoneGoogle added a feature to Google Earth allowing users to rewrite satellite imagery with text prompts — essentially Photoshop for the actual planet — then quietly removed it within 24 hours. One early example showed Pompeii reimagined with a Nano Banana 2, which is either a bold archaeological hypothesis or a sign that nobody ran this past legal. The tool raised obvious concerns about fabricating real locations for misinformation. Google, to its credit, counted to one and pulled it. Source: The Verge
Snapchat Decides That Content Made by Humans Should, Occasionally, Be Seen by HumansSnapchat has updated its Spotlight recommendation algorithm to exclude fully AI-generated videos, meaning only content made by actual people will surface in its main discovery feed. This is notable less for what it does and more for what it quietly admits: that the platform had, until now, been cheerfully amplifying AI slop alongside everything else. A small stand against the content-farm apocalypse. Whether it holds is another matter entirely. Source: TechCrunch
Science Will Publish More Papers, Understand Less, New Study FindsA modelling study published in Nature predicts that scientists using large language models will produce more research, but of lower quality — more papers, less rigour. The culprit is familiar: publish-or-perish incentives, now turbocharged by AI that makes writing faster and thinking optional. The study does not suggest AI makes scientists worse. It suggests AI makes the existing pressure to churn out work considerably more efficient at producing work that shouldn't exist. Source: Nature.com
Jensen Huang Has Opinions About Open-Source AI, and He'd Like You to Hear ThemNvidia's Jensen Huang has made a rare public case for open-source AI, arguing the US should back accessible models rather than ceding ground entirely to the largest closed systems. This is a principled stance that also happens to benefit Nvidia, whose chips power open and closed models alike — an elegant position to occupy. Whether Washington listens is unclear, but when the man who supplies the shovels in a gold rush tells you which mine to dig, it's worth at least reading the map. Source: Digitimes
As always, the AI is doing its best. That's what concerns us.
**
|