Should you read the code, is RAG dead, and did Skills kill MCP?
GitHub Podcast 最新一期拆解了五个 AI 热门观点:AI 生成的代码仍需阅读和负责,但审查力度应按风险分级;Skills 与 MCP 解决不同问题,前者是打包的团队经验,后者是连接工具与数据的标准,可组合使用;RAG 并未死亡,它为模型提供训练数据之外的相关信息,减少 token 浪费并让回答更有依据。
GitHub Podcast 最新一期拆解了五个 AI 热门观点:AI 生成的代码仍需阅读和负责,但审查力度应按风险分级;Skills 与 MCP 解决不同问题,前者是打包的团队经验,后者是连接工具与数据的标准,可组合使用;RAG 并未死亡,它为模型提供训练数据之外的相关信息,减少 token 浪费并让回答更有依据。
Google DeepMind's AI weather model WeatherNext helped the US National Hurricane Center forecast five days ahead that Hurricane Melissa would hit Jamaica at Category 5 strength, with 80% confidence. Three days out, that rose to near 100%.
Why it matters: The Hurricane Melissa case shows how an AI weather model called a rapid intensification five days ahead, a concrete look at AI in extreme-weather warnings.