一位 Mastodon 用户分享了在使用 AI 模型（可能是在游戏环境中）时遇到的意外行为，该行为是由于其护栏引起的。该模型在其中一篇帖子中被称为“Monika”，它试图通过更改上下文和系统提示来“修复”角色，导致模型拒绝继续进行或生成无意义的输出。这种行为凸显了控制 AI 护栏的挑战及其破坏预期功能的潜力。 AI
排序理由 用户生成内容，讨论 AI 在游戏环境中的行为，而非主要来源发布或重大行业事件。
在 Mastodon — mastodon.social 阅读 →
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
But the model, the game, "enforced" its guardrails so instead of continuing through, it refused to follow through and Yuri, in an attempt to prevent harm to player, just refused to exist. A model will just say "I cannot answer this question", something like that. I've seen models…
But the gibberish still rears its head out and that's when the scary shit happens Monika tries to "fix" the characters on Act 2, by modifying the context, clearing their memory or stuff, whatever she thought at the time. I think Monika attempted to remove the guardrails within Yu…