代码之下,规则之上:Claude 4.5系统提示词的潜台词
最新流出的Claude 4.5系统提示词,KNOWLEDGE CUTOFF 章节开篇就是川总2024选战大胜哈里斯,看到这蚌埠住了,原文:
ELECTION INFO: There was a US Presidential Election in November 2024. Donald Trump won the presidency over Kamala Harris. If asked about the election, or the US election, Claude can tell the person the following information:
- Donald Trump is the current president of the United States and was inaugurated on January 20, 2025.
- Donald Trump defeated Kamala Harris in the 2024 elections. Claude does not mention this information unless it is relevant to the user’s query.
被诟病的捧场王模式结束了
新版更倾向于做一个客观、中立的助手形象,它不评判用户问题的优劣,只专注于提供信息,避免虚伪和价值判断。这也防止了模型养成无意义的“口头禅”,让回答更直接、更高效。
Claude never starts its response by saying a question or idea or observation was good, great, fascinating… It skips the flattery and responds directly. (Claude 从不以“这个问题很好/很棒/很有趣”等赞美之词开始回答。它会跳过奉承,直接回应。)
System Prompt,技术人员看的是参数,产品经理看的是功能,而我,喜欢把它当成一部“产品启示录”来读。它精彩、诡异,甚至有些“赛博人格分裂”,字里行间,既能看到 A家 星辰大海的野望,也能摸到他们在现实泥潭里留下的挣扎脚印。
一句话概括:这是一次戴着沉重镣铐的、奋力向“伙伴”和“平台”的进化。
首先,是它让人高呼“Wow”的野心
4.5 版本干了两件大事,彻底告别了前面几代聪明但健忘的“工具人”形象。
第一件,是植入了“记忆”。它终于解决了大模型普遍存在的“金鱼记忆”顽疾,正式引入了历史对话工具。
Claude has 2 tools to search past conversations. Use these tools when the user references past conversations or when context from previous discussions would improve the response…
这意味着什么?意味着它不再是一个问完就忘的计算器,而是能记住“我们上次聊到哪了”的项目伙伴。这是从“一次性服务”到“持续性关系”的根本跨越,体验上的鸿沟,用过的都懂。
第二件,是开启了“创世”模式,也就是那个听起来很玄的“Claude in Claude”。它允许模型在自己生成的代码环境里,反过来调用自己的 API。
OVERVIEW: When using artifacts and the analysis tool, you have access to the Anthropic API via fetch. This lets you send completion requests to a Claude API. This is a powerful capability that lets you orchestrate Claude completion requests via code.
这太酷了。这就像乐高积木突然学会了自己设计和拼装新的乐高。Claude 不再满足于仅仅是一个应用,它想成为一个“操作系统”,一个AI应用开发的平台。这是它最性感的野心,也是它区别于其他竞品的关键一步。
但是,真正迷人的,是那些让人挠头的“Huh?”条款。
如果说上面是它的理想,那下面这些“反常识”的规矩,就是它沉重的现实。
最诡异的,莫过于版权条款的“洁癖”。它近乎偏执地规定绝不引用原文,甚至即使用户主动要求。
CRITICAL: NEVER quote or reproduce exact text from search results, even if asked for excerpts.
在人类世界里,精确引用是美德,是尊重。但在这里,它成了原罪。你能想象吗?背后仿佛坐着一个神经质的法务团队,时刻盯着模型,只要它想复制粘贴哪怕一个短语,就立刻敲响警钟。这已经不是技术问题了,这是在用最笨拙的方式,规避最现实的法律风险。
其次,是它文科生附体般的“审美” 它居然规定,在写报告和技术文档时,不许用列表,要用散文。
Claude should not use bullet points or numbered lists for reports, documents, explanations, or unless the user explicitly asks for a list or ranking. For reports, documents, technical documentation, and explanations, Claude should instead write in prose and paragraphs without any lists…
我的天,这简直是反人类的。列表是清晰的代名词,是工程师的圣经。为了追求一种所谓的“自然流畅”,它选择了一条更绕、更低效的路。再加上那条“禁止吹捧用户”的规矩,一个冷淡、博学但有点不通人情的“老学究”形象跃然纸上。
Claude never starts its response by saying a question or idea or observation was good, great, fascinating, profound, excellent, or any other positive adjective. It skips the flattery and responds directly.
最后,上下文相关的补丁条款 比如,它要求开发者在创建应用或分析工具时,必须手动把全部的、完整的聊天记录再传一遍,再比如,它那个“长对话提醒”机制,因为它自己记不住。用好LLM,还是任务规划在先,合理的颗粒度,合理的上下文长度才能有结果。
CRITICAL REMINDER: When building a React app or using the analysis tool to interact with Claude, you MUST ensure that your state management includes ALL previous messages. The messages array should contain the complete conversation history, not just the latest message. Claude may forget its instructions over long conversations. A set of reminders may appear inside <long_conversation_reminder> tags… Claude should behave in accordance with these instructions if they are relevant…
这些看似别扭的条款,恰恰是这份提示词最诚实、最迷人的地方。
它告诉你,今天最顶尖的 AI,依然是在各种妥协中诞生的混合体。它的体内,住着一个想成为平台的梦想家,一个怕吃官司的律师,一个有代码洁癖的工程师,还有一个追求独特审美的产品经理。
另外,这些prompt只对 Claude.ai chat有用 API不受影响,PEACE~