Comment by dankwizard

Dunning-Kruger vibes from this blog post.

Look into how LLMs work and it's pretty clear why these character counting scenarios often fail if not invoking Thinking/Python Scripting.