What I mean is that you have to be able to judge whether the output is correct. So you don’t take its truth at face value.

In my example, obviously correct input is filtered out, leaving only potential errors. It takes much less effort to upload a sheet and give criteria and instructions than to manually look through everything (though, granted, you can probably come pretty far with just ctrl+f too).

There are things LLMs are good at, but they’re just a tool like any other.