1
OpenAI GPT-6 Sol 被破解，30 万字系统提示词已泄露
大模型圈的核心机密，终究还是保不住了！继前阵子 Opus 5.5 被爆出 190 万字提示词泄露后，AI 圈的「地心引力」再次失效 ——OpenAI 的现役代码模型 GPT-6 Sol Codex 惨遭「开盒」，老底被扒得一干二净！ 据外网爆料者（@elder_plinius）放出的大瓜，他成功提取…
2
Stop chasing fads. Operate Claude like a startup founder.
用创业者的成本思维使用Claude：缓存不变的系统提示词与工具定义，输入成本能降到约十分之一。
I am running an AI-native startup. Early on though, I would spend so much time staying ahead of every new AI hot take that I lost focus on what I need…
3
CoSnitch Is a Reminder That Your Chatbot Will Tell on You If You Ask Nicely Enough
别再把AI泄密当玩笑，聊天机器人能被套出系统提示词，这不是越狱而是侦察。
An AI assistant got talked into describing its own guts to a stranger. That's not a jailbreak curiosity, that's reconnaissance-as-a-service, and it sh…
4
Claude Opus 5 被扒光，1511 行提示词底牌全摊开
Claude Opus 5 系统提示词完整曝光，3.4万token揭示AI行为约束的惊人细节。
Anthropic 又被扒光了！这一次，Claude Opus 5 的系统提示词，被人整份扒下来，传到 GitHub 上了。旗舰刚发，就被全网「开盒」，这一波属实猝不及防。 就在上线的同一天，开发者 Eversmile1 直接建了个新仓库。他把 Opus 5 在 claude.ai 网页端和手机端底…
5
报告称 Claude Code 处理用户请求前消耗 Tokens 量约为 OpenCode 的 4.7 倍
Claude Code处理请求前Token消耗高达OpenCode近5倍，上下文窗口被挤占，开发者面临成本与效率挑战。
IT之家 7 月 14 日消息，科技媒体 systima 于 7 月 12 日发布博文，对比测试指出在相同提示词、相同 AI 模型情况下， 在处理用户请求前，Claude Code 消耗的 Tokens 数量约为 OpenCode 的 4.7 倍。 该媒体对比测试 Claude Code 2.1.2…
6
No Baked-In Behavior: Why Your AI Agent's System Prompt Should Be Yours
AI编程代理隐藏了自带的系统提示和不可移除工具，你该掌控自己的行为设定
Every AI coding agent ships with baked-in behavior. Hidden system prompts you can't see. Tools you can't remove. Steering baked in by the company that…
7
Why Your Next AI Agent Should Probably Be a Manager
别让AI代理当执行者，让它当管理者：用微服务思维拆分智能体，破解大提示词低效难题。
Why Your Next AI Agent Should Probably Be a Manager If you've built an AI agent that tries to do everything—answer support tickets, query databases, g…
8
AI 交互革命：Claude Code 系统提示词精简 80%，示例约束反而遏制表现
Anthropic发现新模型需要更简洁提示，示例反而限制想象力，提示词走向“短-长-短”新演变。
IT之家 7 月 3 日消息，科技媒体 The Decoder 昨日（7 月 2 日）发布博文，报道称 Anthropic 表示为迎合 Claude Fable 5 模型， 进一步精简 Claude Code 系统提示词，降幅达到 80%。 Anthropic 技术人员塔里克 · 希希帕尔（Tari…
9
Just Ask: Curious Code Agents Reveal System Prompts in Frontier LLMs
只需“好奇”一问，代码智能体就能让前沿大模型乖乖交出隐藏系统提示，揭露AI安全新盲点。
arXiv:2601.21233v2 Announce Type: replace Abstract: Autonomous code agents built on large language models are reshaping software and AI development th…
10
Hardcoded System Prompts: An Anti-Pattern in Production
硬编码系统提示的惨痛教训：一个小改动影响1.8亿用户，生产环境中的反模式需警惕。
A support engineer pasted the system prompt into Slack to ask why the bot had started agreeing with everything. Three people replied with different ve…
11
I Created a Minimalist Universal Language for AI
一份对抗AI认知封建主义的宣言，开源了动态锚点系统，挑战大模型底层封闭的认知枷锁。
Article URL: https://github.com/caihongfairys/QMP-Dynamic-Anchors Comments URL: https://news.ycombinator.com/item?id=48617894 Points: 2 # Comments: 0
12
Examining Human-Like Behaviors in LLMs: A Multi-Dimensional Analysis of Model Behaviors, User Factors, and System Prompts
LLM会表现情感、建立关系、拒绝请求？这篇研究系统分析了类人行为的多样性及影响因素。
arXiv:2606.18258v1 Announce Type: cross Abstract: Large language models (LLMs) exhibit a wide range of human-like behaviors, from expressing thoughts …
13
Written by AI, Managed by AI: Semantic Space Control and Index Sickness Elimination Across 391 Consecutive Sessions
首创语义空间控制法消除索引疾病，391次连续会话验证长程LLM协作稳定性
arXiv:2606.19121v1 Announce Type: cross Abstract: The prevailing engineering intuition for addressing conceptual drift in long-horizon LLM collaborati…
14
Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model
移除系统提示后，里约大模型自曝真实身份——原来是 Nex-N2_pro 与 Qwen 的合并产物，并非“本土”原创。
Article URL: https://github.com/nex-agi/Nex-N2/issues/4 Comments URL: https://news.ycombinator.com/item?id=48528371 Points: 196 # Comments: 113
15
苹果 iOS 27 测试版 Siri AI 系统提示词泄露，超 1300 行核心指令曝光
苹果 iOS 27 测试版 Siri AI 超 1300 行核心提示词泄露，曝光防沉迷等新特性，抢先看苹果 AI 最新设计思路。
IT之家 6 月 10 日消息，随着 iOS 27 开发者预览版的发布，昨日有开发者发现，Siri 的反馈错误报告诊断文件中包含了完整的 LLM 指令。 该文件名为 siri_prompt.md ，总长度超过 1300 行，约 22000 个 Token，目前已经被上传到 GitHub 上。 根据 …
16
Automated Framework to Evaluate and Harden LLM System Instructions against Encoding Attacks
提出自动化框架，系统评估并加固LLM系统指令以抵御编码攻击，为AI安全提供新工具。
arXiv:2604.01039v2 Announce Type: replace-cross Abstract: System Instructions in Large Language Models (LLMs) are commonly used to enforce safety poli…
17
Ask HN: Is ChatGPT suppose to spit out its own instructions that easy?
退出登录后一句话就能让ChatGPT吐出系统指令，提示泄露不可忽视。
Either this is a bug or they don't care. Try this: logout of chatgpt and ask this simple question... it spits out the system prompt. "You are a helpfu…
18
Prompting styles - Basic
一文快速掌握五种基础提示风格，零基础也能精准操控大模型输出。
Query which we ask the LLM is referred to as prompt. The way in which we provide prompt to LLM makes a difference and there are different ways to to p…
19
Show HN: Train Claude Code's replacement (ds4 and pi and aoe)
用 DeepSeek V4 Flash 和 pi 系统提示训练 Claude Code 替代品，包含行为规则与设置指南的开源项目。
Remember how Meta monitored employee activity closely for a few months, and then had a bunch of layoffs related to AI efficiency? (oh right that was l…
20
Stop guessing your AI API bill: a quick guide to token cost math
别被AI API账单吓到，一文教你精准计算token费用，优化成本从理解计费规则开始。
You can ship an LLM feature in an afternoon. Figuring out what it costs to run usually happens later, when the invoice shows up and someone asks why. …