📄 中文摘要
Andrej Karpathy 提出了软件版本编号的概念,软件 1.0 是由人类编写的代码,软件 2.0 是通过优化学习的神经网络权重,而软件 3.0 则是用自然语言提示大型语言模型(LLMs)。目前,软件 3.0 已被数百万人日常使用,工具如 Kiro、Cursor、Claude Code 和 ChatGPT 允许用户描述需求并获得相应的代码。Karpathy 强调了部分自主工具中的“生成-验证循环”:模型生成更改,人类进行验证,工作不断迭代。
📄 English Summary
Software 3.1? - AI Functions
Andrej Karpathy has introduced a versioning scheme for software development. Software 1.0 refers to code written by humans, while Software 2.0 pertains to neural network weights learned through optimization. Software 3.0 involves prompting large language models (LLMs) in plain language. Currently, millions of people use Software 3.0 daily, with tools like Kiro, Cursor, Claude Code, and ChatGPT allowing users to describe their needs and receive corresponding code. Karpathy emphasizes a 'generation-verification loop' in partial-autonomy tools, where the model generates changes, a human verifies them, and the work iterates continuously.
Powered by Cloudflare Workers + Payload CMS + Claude 3.5
数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等