Chat templates give LLMs a self‑referential twist
Self‑referential mode activated: A new chat template forces large language models to talk about themselves, turning every reply into a meta‑commentary. The change is already being discussed on Hacker News.
Why it mattersIf you build AI assistants, expect a quirky edge case in prompt handling.





