推文指出,AI开发中的“厂商锁定”概念并不牢固。作者因不满首个选择的错误信息,在不到30分钟内就完全迁移了一个全栈AI应用,该应用包含流式、队列、竞技模式及媒体生成等语义。作者引用并回应了realGeorgeHotz关于“需要世界模型以实现全保真度”的观点,认为在实践中,可以通过“随心构建”可拆解的完整项目,并利用技能和审查来引导大语言模型,就能在无需更多额外努力的情况下高效地进行平行开发,成果显著。
concept of vendor lock-in is kinda fake when i completely migrate a fullstack ai app in <30 mins just bc i didnt like how my first choice had nasty error messages
(with lots of subtle streaming/queuing/arena mode/media generation semantic)
@realGeorgeHotz is right that we need world models for full fidelity, but try just vibing entire projects that are breakable toys and learn to use skills/reviews to keep llms from going off the rails and its amazing how much you can get done in parallel without further effort