Benchmarking AI limits: Microsoft's DELEGATE-52 benchmark shows most AI models falter in extended workflows, corrupting significant portions of content. Domain-specific success: Python-based, highly ...