介绍

本章介绍如何使补全结果更加可靠，以及如何通过检查来确保补全结果的可靠性。

在一定程度上，前面介绍的大部分技术都与提高补全准确度及可靠性有关，特别是自洽性^{1Wang, X., Wei, J., Schuurmans, D., Le, Q., Chi, E., Narang, S., Chowdhery, A., & Zhou, D. (2022). Self-Consistency Improves Chain of Thought Reasoning in Language Models.} 。然而，除了基本提示策略之外，还有许多其他技术可以用于提高可靠性。

LLMs 存在各种问题，包括幻象^{2Ye, X., & Durrett, G. (2022). The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning.} 、采用 CoT 方法的错误解释^{2Ye, X., & Durrett, G. (2022). The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning.} ，以及多种偏差，包括多数标签偏差、近期偏差和常见令牌偏差^{3Zhao, T. Z., Wallace, E., Feng, S., Klein, D., & Singh, S. (2021). Calibrate Before Use: Improving Few-Shot Performance of Language Models.} 。此外，在处理敏感话题时，zero-shot 思维链可能会产生特别的偏差^{4Shaikh, O., Zhang, H., Held, W., Bernstein, M., & Yang, D. (2022). On Second Thought, Let’s Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning.} 。

一些常见的解决方案包括使用校准器消除先验偏差，使用验证器对补全结果进行评分，以及在补全结果中增进多样性。

校准大语言模型

🟢 提示去偏差

🟦 提示多样性

🟦 Prompt Ensembling

🟦 大语言模型自我评估

🟦 Math

Footnotes

1. Wang, X., Wei, J., Schuurmans, D., Le, Q., Chi, E., Narang, S., Chowdhery, A., & Zhou, D. (2022). Self-Consistency Improves Chain of Thought Reasoning in Language Models. ↩

2. Ye, X., & Durrett, G. (2022). The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning. ↩ ↩²

3. Zhao, T. Z., Wallace, E., Feng, S., Klein, D., & Singh, S. (2021). Calibrate Before Use: Improving Few-Shot Performance of Language Models. ↩

4. Shaikh, O., Zhang, H., Held, W., Bernstein, M., & Yang, D. (2022). On Second Thought, Let’s Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning. ↩