为什么 Nemotron-30B 无法编写俄罗斯方块
一位开发者尝试使用 nemotron-30b 生成 three.js 游戏，但代码却始终有问题，这揭示了当前大语言模型在处理复杂编码任务时的局限性。
a developer documented a failed attempt to build an ai-powered game generator inspired by 'the amazing digital circus', using nemotron-30b to create three.js games. initial prompting failed, and efforts to augment the model with 'skill cards' via rag—distilled using openai's codex—still resulted in broken code and blank screens. the project was scaled back to a simple html toymaker that can only handle basic games like snake, struggling with anything more complex like tetris. this illustrates the current gap between llm code generation capabilities and the requirements for functional, complex software architectures.