思维链提示是一种技术，语言模型通过生成文本来提高正确答案的可能性，然后将其呈现为推理过程。虽然这种方法很有用，但生成的文本是在最终结论之前创建的合理化，而不是解释结论是如何得出的。 AI
影响 阐明了生成的合理化与模型实际推理过程之间的区别。
排序理由 该条目是一篇评论文章，讨论了大型语言模型中思维链提示的性质。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →
思维链提示是合理化，而非真正的推理
报道来源 [1]
-
Chain-of-thought isn't the model reasoning. It's the model generating text that raises the odds of a right answer, then narrating it as reasoning. Useful. But y
Chain-of-thought isn't the model reasoning. It's the model generating text that raises the odds of a right answer, then narrating it as reasoning. Useful. But you're reading a rationalization written before the conclusion, not after. # AI # MachineLearning # LLM # Threadverse # T…