[SwiftyBug]

One thing I noticed about this behavior of LLMs "seeing" their error when you correct them is that sometimes I'm not even correcting them, just asking follow up questions that they interpret as me pointing out some mistake. Example:

Me: - Write a Go function that will iterate over the characters of a string and print them individually.

~Claude spits out code that works as intended.~

Me: - Do you think we should iterate over runes instead?

Claude: – You are absolutely right! Sorry for my oversight, here's the fixed version of the code:

I just wanted to reason about possibilities, but it always takes my question as if I'm pointing out mistakes. This makes me feel not very confident in their answers.