University of Waterloo study finds AI models fail software output tasks 25% of the time
New research from the University of Waterloo indicates that AI models fail to deliver reliable structured output 25% of the time. While we monitor the efficiency gains offered by generative AI in professional workflows, the StructEval study highlights a reliability gap.