因子分解诅咒：你预测哪些 token 是反转诅咒及更多现象的根本原因
The Factorization Curse: Which Tokens You Predict Underlie the Reversal Curse and More
AI总结：
本文将反转诅咒重新定义为因子分解诅咒，通过受控实验和 WikiReversal 设置证明其源于大语言模型的下一 token 预测目标，并发现与因子分解无关的目标可显著缓解该问题。
摘要：
当今最优秀的语言模型仍然面临幻觉问题：生成事实错误的内容，这阻碍了它们可靠地检索训练期间见过的信息。反转诅咒——即模型在以不同于训练时遇到的顺序进行探测时无法回忆信息——正是信息检索中这一问题的典型体现。我们将反转诅咒重新表述为因子分解诅咒——即模型无法在不同的因子分解方式下学习相同的联合分布。通过一系列控制实验，并逐步提高真实程度，包括我们引入的 WikiReversal 设置（该设置紧密模拟知识密集型微调任务），我们发现因子分解诅咒是当前流行的大语言模型所使用的下一 token 预测目标所固有的缺陷。此外，我们证明了可靠的信息检索无法通过扩大规模、反转 token、甚至朴素的双向注意力训练来解决。因此，在专门数据上进行微调的各种方法在下游任务上必然会产生好坏参半的结果，除非模型已经见过正确的 token 序列。在五个复杂度不同的任务上，我们的结果揭示了一条有前景的前进道路：与因子分解无关的目标可以显著缓解反转诅咒，并暗示了改进的知识存储和规划能力。
英文摘要：
Today's best language models still struggle with hallucinations: factually incorrect generations, which impede their ability to reliably retrieve information seen during training. The reversal curse, where models cannot recall information when probed in a different order than was encountered during training, exemplifies this in information retrieval. We reframe the reversal curse as a factorization curse - a failure of models to learn the same joint distribution under different factorizations. Through a series of controlled experiments with increasing levels of realism including WikiReversal, a setting we introduce to closely simulate a knowledge intensive finetuning task, we find that the factorization curse is an inherent failure of the next-token prediction objective used in popular large language models. Moreover, we demonstrate reliable information retrieval cannot be solved with scale, reversed tokens, or even naive bidirectional-attention training. Consequently, various approaches to finetuning on specialized data would necessarily provide mixed results on downstream tasks, unless the model has already seen the right sequence of tokens. Across five tasks of varying levels of complexity, our results uncover a promising path forward: factorization-agnostic objectives can significantly mitigate the reversal curse and hint at improved knowledge storage and planning capabilities.
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。
