The hype, based on benchmarks that don't measure real world performance, leads decision makers to overestimate capabilities.

The danger is not super intelligence. The danger is human idiots not understanding what harm insufficiently intelligent ML/AI does in settings like education, government, etc