KK Blog

Menu

Archives

2026 / 07

"Expand an LLM Tokenizer Without Starting Over"

2026-07-22

最佳实践 #LLM#Tokenizer#BPE#Model Migration#Multilingual AI#Inference Performance#Pretrained Models
"A practical contract for expanding a pretrained LLM tokenizer in place while preserving token identity, model quality, and real inference performance
"Agent Skills 跨工具兼容：四层移植边界"

2026-07-22

最佳实践 #Agent Skills#SKILL.md#OpenAI Codex#Claude Code#Gemini CLI#GitHub Copilot#AI 编程#Harness Engineering
用内容、发现、调用与运行时四层模型，判断 SKILL.md 在 Codex、Claude Code、Gemini CLI 和 GitHub Copilot 之间到底能移植什么。
"Agent Skills Compatibility Across Four Coding Agents"

2026-07-22

最佳实践 #Agent Skills#SKILL.md#OpenAI Codex#Claude Code#Gemini CLI#GitHub Copilot#AI Coding Agents#Harness Engineering
A four-layer compatibility model for moving SKILL.md workflows across Codex, Claude Code, Gemini CLI, and GitHub Copilot without confusing shared synt
"AI Coding Agent 为什么需要安装前信任边界"

2026-07-22

最佳实践 #AI Coding Agent#软件供应链#安装前安全#包管理器#沙箱#依赖安全#Harness Engineering
AI Coding Agent 可能在代码审查前执行依赖。本文给出验证包名、来源、版本、哈希和安装脚本的 fail-closed 准入方案。
"AI Coding Agents Need a Pre-Install Trust Boundary"

2026-07-22

最佳实践 #AI Coding Agents#Software Supply Chain#Pre-Install Security#Package Managers#Sandbox#Dependency Security#Harness Engineering
AI coding agents may execute dependencies before review. Build a fail-closed pre-install gate for package identity, source, version, hashes, and scrip
Singular Bank 金融 AI 实践：ChatGPT、Codex 与可验证工作流

2026-07-21

最佳实践 #OpenAI#Singular Bank#ChatGPT#Codex#金融AI#企业AI#银行生产力#可验证性#人机协作
从 Singular Bank 的客户案例看金融 AI：速度只是第一层价值，批准数据源、结构化工作流、输出留痕与人类判断共同构成可验证闭环。
Singular Bank AI Workflows with ChatGPT and Codex for Verifiable Banker Productivity

2026-07-21

最佳实践 #OpenAI#Singular Bank#ChatGPT#Codex#Financial AI#Enterprise AI#Banker Productivity#Verification#Human in the Loop
How Singular Bank reports saving bankers 60–90 minutes daily with ChatGPT and Codex, and why verifiable workflows matter in finance.
"AI 建议的信心陷阱：错误答案如何压缩我不知道的空间"

2026-07-20

解读 #AI建议#自动化偏差#批判性思维#决策#AI信心#人机协作#可验证性
"五项实验显示，错误 AI 建议会降低准确率并抬高信心。真正可执行的应对，是把独立判断和外部核验做进工作流。"
"The AI Advice Confidence Trap: Why Wrong Answers Suppress 'I Don't Know'"

2026-07-20

解读 #AI Advice#Automation Bias#Critical Thinking#Decision Making#AI Confidence#Human AI Collaboration#Verification
"A five-experiment preprint shows how wrong AI advice can reduce accuracy and inflate confidence. The practical answer is a three-stage verification l
"AI Agent 什么时候需要自己的 DSL：一套最小决策框架"

2026-07-20

最佳实践 #AI Agent#领域特定语言#LLM#Agent可靠性#DSL#可验证性#软件架构
"判断 AI Agent 何时需要领域特定语言，何时 JSON Schema 已经够用，并用最小 DSL 架构建立可验证的执行边界。"
Previous Next