A new benchmark reveals that even the best AI models fail at realistic knowledge work, solving only 3% of tasks completely.

This highlights the current limitations of AI in complex, multi-step workflows.

AI-generated: summaries written by AI from the linked sources. How we use AI