Staff Correspondent

How do you know if an AI model is truly smart, or just good at passing tests? A sweeping new study from the University of Oxford  has revealed that many of the world’s most popular AI benchmarks may not accurately measure what they claim. The research, titled “Measuring What Matters: Construct Validity in Large Language Model Benchmarks,”  analyzed 445  benchmarking papers from top AI and machine learning conferences, and the findings are raising eyebrows across the industry. A Benchmark...

AI might be the Future- but that Future is coming with a Bigger Electricity Bill! Every time an AI model trains, a chatbot answers, or a tool generates an image, somewhere a data center is pulling megawatts of electricity to make it happen. Now, with global energy prices soaring, that hidden cost is coming into focus. The tech world’s biggest challenge isn’t just building more intelligent AI anymore; it’s finding the power to keep it running. The Growing Strain on AI Infrastructure AI doesn’t...

Welcome to the October 2025 edition of our regulatory newsletter, featuring landmark reforms in financial supervision and prudential policy worldwide. This month’s highlights: the Federal Reserve’s push for transparency in bank stress tests, India’s adoption of the global IFRS 9 credit loss model, new monetary signals from the Fed, major EU and UK rule consultations, pivotal Basel Committee and Singapore MAS amendments, and advances in fintech governance. Regulatory authorities responded to...