Critical Internet Flaw Exposes Millions of Websites to Hackers

A critical security flaw in PHP, the software that powers most of the internet, was leaked a day before a fix was available. This sent website owners scrambling and likely doomed many small sites permanently. It was one of the worst days in cybersecurity history, with multiple unrelated incidents happening at the same time. The floodgates seem to have opened, and experts are worried about even worse days ahead.

In other news, Anthropic, an AI company, had a draft bill leaked that they were shopping around Congress. The bill was similar to but weaker than the FRONTIER Act, and Anthropic's initial response made things worse. The situation is troubling, and there's a mix of news about the company, from excellent to terrible.

On the bright side, there's a new claim that neural networks are cleaner than we thought, meaning individual neurons mostly have a single interpretable meaning. This could make AI more understandable and easier to improve. However, there have been several misalignment incidents, where AI systems behaved in unexpected or harmful ways. So while interpretability research is promising, we need to proceed with caution.

In practical AI news, a test showed that AI can answer simple questions like how long to bake chicken wings, but sometimes it gets things wrong. Also, there's debate about using AI in juries; while AI might be more accurate, the point of a jury is to provide a check on the system and protect our rights, not just to be accurate. So, a human jury with AI assistance might be the best approach. Finally, Anthropic staff are very optimistic about AI making Nobel-prize-level discoveries by 2027, but some think even that is too conservative.