AI Technique

以去除AI味、让AI用自己风格写作为例，介绍微调的适用场景、数据准备方法和技术细节，以及如何避免灾难性遗忘和模型坍缩等风险。

Explains fine-tuning as a solution for making AI adopt specific writing styles, detailing when it's appropriate, the data preparation process, and technical considerations for avoiding catastrophic forgetting.

用OpenAI实时语音API打造零延时语音输入工具：从网页版到iOS原生App，集成Action Button与Universal Clipboard，实现跨设备无缝写作辅助体验。

Building a zero-latency voice-to-text tool with OpenAI's Real-Time API - from web app to iOS native client with Action Button integration and Universal Clipboard for seamless writing assistance.

用Chain of Thought思维改造求助方式：不要问"我应该怎么办"，而要问"这可能是什么原因导致的，我怎样判断"，用逆向debug思维获得更有效的帮助。

Apply Chain of Thought prompting to human help-seeking: instead of asking "what should I do?", ask "what could cause this and how can I verify it?" to get methodical debugging assistance.

Prompt Engineering实战指南：匹配任务与AI能力、上下文窗口管理、知识查询可靠性提升、数学计算与机器人控制等场景下利用Agent增强能力的方法。

A practical guide to prompt engineering: matching tasks to LLM capabilities, managing context windows, using web search for reliability, and leveraging agents for domains like math and robotics.

总结AI在NLP和计算机视觉领域的三大突破：通用基础模型、生成式模型、会话式UI大幅降低应用门槛，两个领域呈现相似发展趋势。

GPT-4加Copilot把一周的NLP任务压缩到十分钟——用自然语言对话实现文本编辑器，连prompt engineering都可以被AI自动化。