DELETE FROM users WHERE location = 'IRAN';. GitHub Gist: instantly share code, notes, and snippets.

Sometimes the people who need invitations most are the ones who always decline them.

Brutalita Sans 是一款实验性字体及其配套的在线字体编辑器。

• 核心特性：允许用户直接在浏览器中编辑该字体。
• 主要功能：编辑后可下载 OTF 格式的字体文件。
• 项目定位：实验性质，结合了字体设计与在线编辑工具。

https://mitchellh.com/writing/libghostty-is-coming

http://edwardpackard.com/wp-content/uploads/2025/09/Nine-Things-I-Learned-in-Ninety-Years.pdf

Ruby Central recently took over a collection of open source projects from their maintainers without their consent.

https://openai.com/index/openai-nvidia-systems-partnership/

Qwen3-omni is a natively end-to-end, omni-modal LLM developed by the Qwen team at Alibaba Cloud, capable of understanding text, audio, images, and video, as well as generating speech in real time. - QwenLM/Qwen3-Omni

https://bbbprograms.org/media/newsroom/decisions/mrbeast-feastables

I tweeted years ago; Boarding houses were made illegal by zoning that enforced single family homes and by rules limiting occupancy, demanding every room have a private bathroom, outlawing shared kitchens, requiring parking spaces for every resident etc. How States and Cities Decimated Americans’ Lowest-Cost Housing Option is an excellent, hard-hitting piece making and extending […]

https://www.nytimes.com/2025/09/23/us/politics/secret-service-sim-cards-servers-un.html

Contribute to humanlayer/advanced-context-engineering-for-coding-agents development by creating an account on GitHub.

# 采样与结构化输出在大语言模型中的概述

大语言模型（LLMs）的输出生成过程主要涉及**采样策略**与**结构化输出控制**两大核心机制。

## 核心概念

1.  **采样**：指从模型输出的概率分布中选取下一个词元的过程，直接影响生成内容的多样性、质量与一致性。
2.  **结构化输出**：指引导模型生成符合特定格式（如JSON、XML、表格、列表等）的内容，以提高输出的可解析性和实用性。

## 关键技术与目的

- **采样方法**：
    - **贪心搜索**：始终选择概率最高的词元，输出确定性高但可能重复单调。
    - **束搜索**：并行探索多个可能序列，平衡质量与多样性。
    - **随机采样**：根据概率分布随机选择词元（如温度采样、Top-K采样、Top-P采样），能生成更多样和富有创造性的文本。
    - **温度参数**：调整概率分布的平滑度，高温增加随机性，低温趋于确定性。

- **结构化输出方法**：
    - **提示工程**：在输入提示中明确指定输出格式和要求。
    - **约束解码**：在生成过程中强制将输出限制在符合特定语法规则或模式的词元序列内。
    - **后处理与解析**：对模型生成的自由文本进行格式清洗或转换。

## 主要应用场景

- **需要精确格式的数据提取与转换**（如生成API响应、数据库查询）。
- **自动化的表单填写或报告生成**。
- **与外部程序或系统集成**，要求输入输出具有严格的数据结构。

## 面临的挑战

- **权衡质量与约束**：过于严格的格式约束可能损害输出内容的自然度和质量。
- **采样与确定性的平衡**：不同的采样策略需要在创造性、准确性和稳定性之间做出权衡。
- **解析的可靠性**：即使指定了格式，模型仍可能生成不符合要求的输出，需要稳健的解析机制。

总结而言，采样策略决定了LLM如何生成文本序列，而结构化输出技术则旨在确保生成结果满足特定的格式规范，两者共同服务于提升LLM输出的可用性与可控性。

x402 is the internet's payment standard. An open standard for internet-native payments that empowers agentic payments at scale. Build a more free and fair internet.

https://platform.leolabs.space/visualization

As a data format, yaml is extremely complicated and it has many footguns. In this post I explain some of those pitfalls by means of an example, and I suggest a few simpler and safer yaml alternatives.

There are over 200 recordings of the Bach Cello Suites to choose from. Read our reviews to give you a better idea of which your may prefer.

Despite a surge in generative AI use across workplaces, most companies are seeing little measurable ROI. One possible reason is because AI tools are being used to produce “workslop”—content that appears polished but lacks real substance, offloading cognitive labor onto coworkers. Research from BetterUp Labs and Stanford found that 41% of workers have encountered such AI-generated output, costing nearly two hours of rework per instance and creating downstream productivity, trust, and collaboration issues. Leaders need to consider how they may be encouraging indiscriminate organizational mandates and offering too little guidance on quality standards. To counteract workslop, leaders should model purposeful AI use, establish clear norms, and encourage a “pilot mindset” that combines high agency with optimism—promoting AI as a collaborative tool, not a shortcut.

Get real-time fall foliage updates and find the best fall color with our interactive map. Plan your fall trip today!

A smarter cd command. Supports all major shells. Contribute to ajeetdsouza/zoxide development by creating an account on GitHub.

核心观点

本文作者认为，在开发实际的命令行工具（CLI）时，Zig 语言比 Rust 更为实用和高效。这主要基于两者在内存管理、开发体验和对“软件安全”定义的不同取向。

关键论据与对比

1. 内存管理机制：

   • 栈与堆：文章首先解释了栈（高效、固定大小、LIFO）和堆（动态分配、可能碎片化）的基本区别，这是理解两者语言设计的基础。
   • Rust 的借用检查器：这是一个强大的编译时内存安全工具，能防止悬垂指针、数据竞争等。但其局限在于：仅在编译时工作，无法修复逻辑错误或糟糕的设计；会带来认知负担，迫使开发者围绕编译器重构代码以满足生命周期和所有权规则，在小工具中显得“过重”。
   • Zig 的分配器与 defer：Zig 提供手动内存管理，但通过结构化的分配器（如 GeneralPurposeAllocator）和 defer 语句进行资源清理。这种方式更直观，允许开发者自由分配、存储指针并修改数据，而无需与生命周期“缠斗”。它将责任置于开发者身上，要求纪律，但换来的是更简单、更易推理的代码。

2. 实践案例：笔记 CLI 工具：

   • 在 Rust 中实现一个解析笔记并构建标签索引的功能时，尝试添加新笔记会与现有引用的可变性冲突，导致与借用检查器的对抗，增加了不必要的复杂性和样板代码（如克隆、使用 Rc/RefCell）。
   • 在 Zig 中，同样的功能实现起来非常直接：用分配器创建列表，存储指向其中元素的指针以建立索引，然后自由修改列表。没有额外的包装器或编译器束缚。

3. 对“软件安全”的再思考：

   • 文章指出，内存安全只是软件安全的一个方面。Rust 侧重于此，但 CLI 工具的安全还包括：可预测的行为（处理畸形输入）、避免崩溃或数据损坏、合理的性能、敏感数据保护和抗攻击能力。
   • Rust 的借用检查器能有效防止特定内存错误，但不能保证逻辑正确、行为可预测或防止其他类型的安全漏洞。因此，将内存安全等同于整体安全是片面的。

4. 开发者体验与哲学：

   • Rust：提供强编译时保证，但开发体验存在“仪式感”（ceremony）和认知开销，代码有时“为编译器服务”。适合大型、并发、长生命周期的系统。
   • Zig：强调开发者友好和务实。它提供清晰的工具（分配器、defer、comptime），要求开发者负责但不过度干预。作者认为，这尊重了开发者的能动性，将他们视为能通过实践学习并改进的“成年人”，而非需要编译器时刻监督的“孩童”。这带来了更佳的开发工效学，让编码过程更愉悦、高效。

结论

对于作者所构建的小型、实用的 CLI 工具而言，Rust 的严格性可能“过犹不及”，导致开发效率降低。而 Zig 凭借其轻量、直接和对开发者的信任，在提供必要内存安全控制的同时，允许更快地迭代和编写易于理解的代码，因此显得更为实用。最终，语言选择应契合工作流和项目类型——在此场景下，Zig 击中了“甜蜜点”。

Alex Moon is an Australian-British software engineer and artist based in the North East of England.

Forty-five Maine prisoners are currently working remote jobs for outside companies. A few are working full-time, earning more than corrections officers. One is making well into the six figures. Here's why Maine's approach to education and work in prison is making waves in the world of corrections.

People should spend less time learning DSA, more time learning testing.

Disney says "thoughtful conversations with Jimmy" led to show's return.

A federal judge has ruled that a nearly complete offshore wind project halted by the Trump administration can resume, dealing President Donald Trump a setback in his ongoing effort to restrict the U.S. offshore wind industry.

Strategy – how to be strategic, and how to be seen as strategic – is one of my ongoing obsessions. Years ago, I read Good Strategy/Bad Strategy, and it’s guided my thinking ever s…

Section 1.  Antifa as a Terrorist Threat.  Antifa is a militarist, anarchist enterprise that explicitly calls for the overthrow of the United States

Astronomers have long relied on supercomputers to simulate the immense structure of the Universe, but a new tool called Effort.jl is changing that. By mimicking the behavior of complex cosmological models, this emulator delivers results with the same accuracy — and sometimes even finer detail — in just minutes on a standard laptop. The breakthrough combines neural networks with clever use of physical knowledge, cutting computation time dramatically while preserving reliability.

Gerrit Code Review

The court of appeals issued an historic fine after 21 of 23 quotes in the lawyer's opening brief were fake. Courts want more AI regulations.

Well, if you are reading this obituary, it looks like I’m dead. WOW, it actually happened … I died of FOMO due to complications of Bulbar ALS. My name is Linda Brossi Murphy, and I was just 60 years old when I died on Sunday evening, September 21, 2025 – way too young!  As the […]

Adventures using AI agents, especially Claude Code, and MCP Servers

https://www.filfre.net/2025/09/choose-your-own-adventure/

Abstract page for arXiv paper 2509.06917: Paper2Agent: Reimagining Research Papers As Interactive and Reliable AI Agents

Absolutely inappropriate. Brendan Carr has got no business weighing in on this."

“In evolution, a maladaptation is a trait that is (or has become) more harmful than helpful, in contrast with an adaptation, which is more helpful than harmful.” I went to Wikipedia to look that up, not a large language model.

https://news.ycombinator.com/item?id=45347914

How MCP Authentication Flaws Enable RCE in Claude Code, Gemini CLI, and More

https://davidlattimore.github.io/posts/2025/09/02/rustforge-wild-performance-tricks.html

Making vector search fast, memory-efficient, and natural in SQLite.

https://www.worksinprogress.news/p/how-to-clean-the-air

https://www.kwikset.com/support/answers/what-does-the-kevo-app-shutdown-mean-to-my-kevo-door-lock

After all you made your apps open source, so I can remove your ads.

Learn how to extract text from audio files locally using Python, Whisper, and other powerful tools. Complete setup guide with practical examples.

OpenDataLoader-PDF 概述

核心定位

OpenDataLoader-PDF 是一个开源 PDF 解析工具，专注于为 AI 提供结构化数据，并自动化 PDF 无障碍处理。它基于 Apache 2.0 许可证发布。

主要功能

1. PDF 数据提取（用于 AI/RAG）

• 高精度解析：在基准测试中综合排名第一（0.907），表格提取准确率高达 0.928。
• 混合模式：结合快速本地处理与 AI 后端，自动优化复杂页面（如复杂表格、扫描件、公式、图表）。
• 多格式输出：支持提取为 Markdown（适合 LLM 分块）、JSON（含每个元素的语义类型和边界框）、HTML 等。
• OCR 支持：内置 OCR，支持 80+ 种语言，可处理 300 DPI 以上的低质量扫描件。
• 本地运行：所有处理均在本地进行，无需上传数据至云端。

2. PDF 无障碍自动化

• 自动标签生成：可将无标签 PDF 自动转换为带结构标签的 Tagged PDF，使其能被屏幕阅读器读取。
• 合规性基础：这是实现 PDF/UA-1/2 等无障碍标准的基础步骤。自动标签功能本身是开源的。
• 企业级扩展：完整的 PDF/UA 导出和可视化编辑器作为企业级功能提供。
• 标准协作：遵循 PDF 协会的 Well-Tagged PDF 规范，并与 veraPDF 验证工具集成。

关键特性与优势

• 确定性 + AI 混合：默认模式下快速本地处理；复杂场景自动调用 AI，精度高且速度快（0.463秒/页）。
• 元素级结构识别：可检测标题、段落、表格、列表、图像、公式等，并提供边界框坐标。
• 阅读顺序处理：采用 XY-Cut++ 算法，正确解析多栏布局。
• AI 安全过滤：自动过滤隐藏文本、页面外内容等提示注入攻击。
• 多语言支持：数字 PDF 开箱即用；扫描件通过混合模式支持多种语言。
• 易于集成：提供 Python、Node.js、Java SDK，并与 LangChain 集成。

使用模式

• 快速模式：适用于标准数字 PDF，纯本地处理。
• 混合模式：适用于复杂文档，需启动本地 AI 后端服务器。
• 标签模式：专门用于生成无障碍 Tagged PDF。

安装与使用

• 要求：Java 11+ 和 Python 3.10+。
• 安装：通过 pip install opendataloader-pdf。
• 基本用法：几行代码即可批量处理 PDF 文件夹，输出指定格式到目录。

基准测试表现

在包含 200 份真实世界 PDF 的测试中，其混合模式在综合准确率、阅读顺序和表格提取方面均领先其他开源及商业解析器。

目标用户与场景

• AI/RAG 开发者：需要从 PDF 中精准提取结构化数据以供模型使用。
• 文档自动化团队：需要大规模将 PDF 转换为无障碍格式。
• 企业与组织：需要遵守 EAA、ADA 等全球无障碍法规。

许可证与未来

• 核心库（包括提取和自动标签）为 Apache 2.0 许可，可免费用于商业用途。
• 企业功能（如 PDF/UA 导出）需要企业许可。
• 路线图包括与 Hancom Data Loader 的企业级 AI 文档分析集成。

Björk talks about an early trip to the dentist, her lifelong interest in nature and technology, and what she learned from last year’s MoMA exhibition.

Permeable materials in homes can retain volatile organic compounds for up to a year

SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks? - scaleapi/SWE-bench_Pro-os

Abstract page for arXiv paper 2503.15443: Are Elites Meritocratic and Efficiency-Seeking? Evidence from MBA Students