Jan Leike· janleike·· 2025-11-22AI 评分31
任务框架决定模型对齐表现
This seems to depend entirely on how you frame the task to the model: If you say "please don't hack" but hacking gets rewarded -> broad m...
AI 导读
这似乎完全取决于你如何向模型定义任务：如果你说“请不要作弊”但作弊获得奖励 -> 广泛不对齐；如果你说“作弊是可以的”且作弊获得奖励 -> 无广泛不对齐
来源：Jan Leike · x.com