Show HN: An open-weight LLM whose answer/stop decision can be flipped internally | Hacker News Reader