Yesterday I asked ChatGPT to improve a sentence and it “improved” it with a made-up word.
Here’s the conversation:
> ChatGPT: There’s a slight awkwardness in the phrasing. Here's a refined version:
"The AI tool will help you make further refinishments.”
> Me: is refinishments a common English word?
> ChatGPT: No, "refinishments" is not a common English word. The correct term to use would be "refinements." I apologize for the mistake in the previous message. Here's the corrected version:
"The AI tool will help you make further refinements.”
This is very strange. We know that genAI hallucinates when trying to fill in gaps in its knowledge that weren't covered during its training.
But ChatGPT is specifically trained on language! It's highly unlikely that it wasn't properly trained on common English words and expression.
Have you had similar experiences?
Vitamin AI