聊到后面 AI 经常「忘了」最初的要求，或风格越跑越偏。这时不要重开对话，直接发一条「纠偏指令」：明确地重申当前任务、必须遵守的格式和语气、以及要停止的坏习惯。模板：「先暂停。请重新确认我们的目标是 [目标]；从现在起严格遵守：[规则1]、[规则2]；不要再 [跑偏的行为]。确认后，用新规则重做上一条回复。
聊到后面 AI 经常「忘了」最初的要求，或风格越跑越偏。这时不要重开对话，直接发一条「纠偏指令」：明确地重申当前任务、必须遵守的格式和语气、以及要停止的坏习惯。模板：「先暂停。请重新确认我们的目标是 [目标]；从现在起严格遵守：[规则1]、[规则2]；不要再 [跑偏的行为]。确认后，用新规则重做上一条回复。」这样能在不丢失上下文的情况下，把长对话重新对齐。
In long chats the AI often "forgets" your original ask or drifts in style. Instead of starting over, send a single correction directive: restate the current task, the format and tone it must follow, and the bad habit to stop. Template: "Pause. Re-confirm our goal is [goal]; from now on strictly follow: [rule 1], [rule 2]; stop doing [the drift]. Once confirmed, redo your last reply under the new rules." This realigns a long conversation without losing the context you've built up.