Microsoft AI models struggle with long, multi-step tasks

Microsoft researchers found AI models introduce significant errors in long, multi-step tasks. Even advanced AI models lost an average of 25 percent of document content over 20 interactions, failing to meet readiness benchmarks in most tested professional domains. The study suggests users must closely monitor AI systems, as tools did not improve performance and errors can occur suddenly, impacting current AI automation investments.