1
Show HN: Muteval – mutation testing for your LLM evals
变异测试一招揪出LLM评测漏洞，Muteval让评估数据更可靠，开源免费直接上手。
Article URL: https://github.com/AshwinUgale/muteval Comments URL: https://news.ycombinator.com/item?id=49388918 Points: 1 # Comments: 0
2
How to Ask the AI: A User Perspective Survey for Large Language Model Prompting
别只顾着问AI，这项用户视角调查告诉你提示词到底该怎么写更有效
arXiv:2608.07494v1 Announce Type: cross Abstract: AI tools like ChatGPT and DeepSeek, powered by Large Language Models (LLMs), allow users to obtain i…
3
I got 30% better AI coding results
实测让 AI 编程效果提升 30% 的代码瘦身法，开源工具 CodeSlimmer 帮你喂给大模型更精准的上下文
Article URL: https://github.com/akasula09/CodeSlimmer Comments URL: https://news.ycombinator.com/item?id=49236555 Points: 2 # Comments: 0
4
Opus 5狂烧6.9亿token做游戏，GPT-5.6用5美元复刻了
单条提示词狂烧6.9亿token做游戏，GPT-5.6却用5美元复刻，AI游戏开发成本战一触即发。
据说GPT-5.6很适合拿来做游戏
5
IA para Autismo No Verbal: Crea tu Tablero CAA con ChatGPT
用ChatGPT生成CAA沟通板，为无语言自闭症者定制简单 pictogramas，免费替代昂贵商业应用。
Llevo años utilizando herramientas de comunicación aumentativa con mi hijo, que tiene autismo no verbal. Hemos probado de todo: desde apps comerciales…
6
OSINT prompts for ChatGPT and Claude: the circular-reporting screen analysts use, in full
用提示词让AI画出引用链，一眼看穿五个来源坍缩成一个的循环报道。
Five sources report the same detail, so you write it up as confirmed. Then you look closer. Two of them cite the third. The third cites a post that ha…
7
Single Canonical Prompts Underestimate LLM Safety's Surface-Form Sensitivity
别被单一提示词骗了！这项研究揭示LLM安全评估常因表面措辞变化而失真，需警惕安全机制的真实覆盖范围。
arXiv:2608.02665v1 Announce Type: cross Abstract: A benchmark score is a measurement instrument, yet most benchmarks read each item at a single canoni…
8
Surgical DevOps – Prevent LLM context drift and regressions
开源工具专治大模型上下文漂移，让AI严格守规矩，代码块神圣不可侵犯。
Article URL: https://github.com/bonushora/surgical-dev-ops/blob/main/README_EN.md Comments URL: https://news.ycombinator.com/item?id=48979553 Points: …
9
OpenAI官方教你8招玩透ChatGPT！
OpenAI官方亲授8个核心技巧，帮你把ChatGPT用到极致，效率翻倍！
10
Why Your ChatGPT Answers Feel Generic (It's Not the Model's Fault)
别再怪AI了，你的提问方式才是回答空洞的元凶——学会给足上下文，对话质量立竿见影。
A while back I was researching a topic I didn't know much about — the kind of casual, late-night "let me just ask the AI a few questions" session. A f…
11
A visual calculator to estimate and compare LLM API pricing
可视化对比各家大模型API定价，跑工作流前先算清token成本，省钱又省心。
Article URL: https://neutraloverdrive.com/tools/token-calculator/ Comments URL: https://news.ycombinator.com/item?id=48817123 Points: 1 # Comments: 0
12
Ask HN: What is your go-to prompt to prove AI can be wrong?
想知道AI何时一本正经地胡说八道？这个HN热帖汇集了各路高手验证大模型错误的“必杀提问”。
Just over-heard one of our seniors trying to explain to a new intern that AI can actually be wrong, no matter how much its answers sound very eloquent…
13
2026 Unslop AI-Written Fiction Contest Results
AI写作竞赛结果出炉，看人类如何用提示词驯服机器，产出不油腻的虚构故事。
Article URL: https://www.hyperstitionai.com/unslop-results Comments URL: https://news.ycombinator.com/item?id=48782890 Points: 21 # Comments: 48
14
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…
15
How to Write DESIGN.md Prose That AI Agents Actually Follow
给AI写设计规范，用“该做/不该做”硬规则，让智能体精准还原界面细节
To write DESIGN.md prose agents follow, describe intent and rules, not just values. A token says a color is a hex; good prose says it is for the prima…
16
Ask HN: What do you do to make LLMs determine
来围观 HN 网友如何调教大模型，获取一手实战思路，让你的 LLM 更靠谱
Comments URL: https://news.ycombinator.com/item?id=48610221 Points: 2 # Comments: 0
17
Why Most Custom AI Skills Never Run (And the One Fix)
AI技能总是静默失效？原因在于触发器太死板。教你用高杠杆改动，让技能随叫随到。
There is a quiet failure mode with AI agent skills that almost everyone hits: you build a custom skill, it looks perfect, and it simply never runs. No…
18
Show HN: open source llm fusion
开源实现OpenRouter的混合智能体方案，融合多模型提升回答质量，无需配置即可上手的实用工具。
OpenRouter's fusion is great. I built this for my own deep research queries. Model fusion is useful for tool calls and complex prompts that require co…
19
You Can't Prompt Your Way Out of a Hard Constraint
用X平台字符限制实例，揭露LLM无法仅靠提示词绕过硬约束，给开发者敲响警钟。
Thursday morning I removed five nodes from my content pipeline. By lunch I understood something about building with language models that eleven failed…
20
Prompting fundamentals
OpenAI官方教程，掌握提示词基础，让ChatGPT输出更精准有用。
Learn prompting fundamentals and how to write clear, effective prompts to get better, more useful responses from ChatGPT.