

You may have led it to look for wrong things. What if you just input the text and ask it to proofread it. Sometimes having a more neutral prompt allows the LLM to lapse more in its predictive nature.
I mean it could have been a human that missed it too. Like I said, when glancing over it myself I didn’t see the right part, I just comprehended it was the other ear. The brain is weird.
Right. I don’t mind people wearing certain red hats. It lets me know right up front what to expect. Reddit had the same dilemma once upon a time where specific subs were banned, and that just forced the roaches to hide in unknown areas where once you knew exactly where to find them all.