seems like GPT systems are the most useful when you can quickly verify the correctness of the output (ex code generation) but else quickly becomes dubious (ex h

seems like GPT systems are the most useful when you can quickly verify the correctness of the output (ex code generation) but else quickly becomes dubious (ex how do you know a summary is correct or useful?)