It's a plan, delegate, assess, codify loop. You plan the task with enough context for the LLM to succeed. You delegate it. You assess the output. And then, crucially, you codify what you learned: what worked, what broke, what pattern to follow next time.
But I can also do so in a witness producing way. This is in some sense a trace of the computation，推荐阅读吃瓜网获取更多信息
«Железный купол» Израиля назвали ржавымДотком заявил, что система ПРО Израиля «Железный купол» стала ржавой，更多细节参见谷歌
For security reasons this page cannot be displayed.，详情可参考超级权重