正式性受青睐：揭示大语言模型在知识冲突数据上的学习偏好
Formality is Favored: Unraveling the Learning Preferences of Large Language Models on Data with Conflicting Knowledge
AI总结：
本研究系统分析大语言模型对知识冲突数据的学习偏好，发现其偏好正式、拼写错误少的文本，该偏好跨模型语言通用且大模型更显著，还可通过调整数据与多数的一致性操控偏好。
摘要：
大语言模型经海量预训练数据训练后，已在众多知识密集型任务中展现出优异性能。然而，预训练数据往往包含具有误导性甚至相互冲突的信息，探究大语言模型（LLMs）在训练过程中如何处理这些噪声数据是一个值得深入研究的问题。本研究系统分析了LLMs对存在知识冲突的数据的学习偏好。我们发现，预训练后的LLMs形成了与人类相似的学习偏好，即更偏好正式文本和拼写错误更少的文本，这使得模型在面对冲突时，对具备此类特征的数据中的知识学习速度更快，也更倾向于采信这些知识。该发现在不同模型和语言间均具有普适性，且在规模更大的模型中表现更为明显。深入分析表明，LLMs倾向于信任具备“与大多数数据一致”特征的数据，并且可以通过操控数据与多数数据的一致程度，来为模型灌输新的偏好、消除旧有偏好。
英文摘要：
Having been trained on massive pretraining data, large language models have shown excellent performance on many knowledge-intensive tasks. However, pretraining data tends to contain misleading and even conflicting information, and it is intriguing to understand how LLMs handle these noisy data during training. In this study, we systematically analyze LLMs' learning preferences for data with conflicting knowledge. We find that pretrained LLMs establish learning preferences similar to humans, i.e., preferences towards formal texts and texts with fewer spelling errors, resulting in faster learning and more favorable treatment of knowledge in data with such features when facing conflicts. This finding is generalizable across models and languages and is more evident in larger models. An in-depth analysis reveals that LLMs tend to trust data with features that signify consistency with the majority of data, and it is possible to instill new preferences and erase old ones by manipulating the degree of consistency with the majority data.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
