Microsoft's MAI models used web data, not just 'clean' sources as claimed

Microsoft's new MAI AI models trained on web data, contradicting earlier claims of using only 'clean, enterprise-grade, commercially licensed' data. The company's technical paper reveals a mix of publicly available web data (like Common Crawl) and licensed human-generated data, not just clean sources as marketed. Microsoft says it uses its own crawler and follows robots.txt rules, but critics argue the 'opt-out' approach puts the burden on website owners. This is seen as a loophole in data ethics.