Next time an AI agent fucks up a task for you, if you ask it (or a different AI) to identify why and suggest ways you can prevent this in the future, I expect the response to be very under-informative, much worse than their ability to help you diagnose communication errors between people.
Seems that’s because there’s significantly more scope to diagnose (rightly or wrongly) communication errors between people, who have motivations, expectations and emotional reactions, belong to cultures and hierarchies and communicate with subtext...
Wouldn’t this be true for any intelligent being, not just LLMs? Almost by definition, you’re disproportionately bad at diagnosing your own mistakes, because otherwise you probably would’ve stopped making them.
Next time an AI agent fucks up a task for you, if you ask it (or a different AI) to identify why and suggest ways you can prevent this in the future, I expect the response to be very under-informative, much worse than their ability to help you diagnose communication errors between people.
Seems that’s because there’s significantly more scope to diagnose (rightly or wrongly) communication errors between people, who have motivations, expectations and emotional reactions, belong to cultures and hierarchies and communicate with subtext...
Wouldn’t this be true for any intelligent being, not just LLMs? Almost by definition, you’re disproportionately bad at diagnosing your own mistakes, because otherwise you probably would’ve stopped making them.