问HN:推理是否仅仅是某种迭代的上下文优化?

2作者: sschmitt9 个月前原帖
上下文 = "你是一个有用的模型。你在初始用户输入后评估自己的输出,持续 N 步" 提示 = "用户发来的提示" 上下文 += 提示 对于 _ 在范围(N)内: 上下文 += 评估_llm(上下文)
查看原文
context = &quot;you are a useful model. you evaluate your own output for N steps after the initial user input&quot;<p>prompt = &quot;here comes a prompt by the user&quot;<p>context += prompt<p>for _ in range(N): context += evaluate_llm(context)