arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2310.15047cs.LGcs.AI
收藏

隐式元学习可能使语言模型更信任可靠来源

Implicit meta-learning may lead language models to trust more reliable sources

University of Cambridge(剑桥大学);Max Planck Institute for Intelligent Systems(马克斯·普朗克智能系统研究所);University of Toronto(多伦多大学)

Dmitrii Krasheninnikov, Egor Krasheninnikov, Bruno Mlodozeniec, Tegan Maharaj, David Krueger

AI总结:

该研究发现大语言模型可通过隐式元学习识别文本有用性指标并调整更新,验证了该现象在多类模型与视觉任务中存在,探究了影响因素及知识存储变化,为AI系统可控性研究提供参考。

摘要:

我们证明大语言模型(LLMs)可能会学习文档有用性的指标,并相应调整自身的参数更新。我们在合成微调数据集中引入随机字符串(“标签”)作为有用性的指标。在该数据集上微调会引发隐式元学习(IML):在后续微调中,模型会更新参数以更多利用被标记为有用的文本。我们对这一现象开展了全面的实证研究,除其他发现外还得出:(i)该现象在预训练LLM、从头训练的LLM中均存在,在视觉任务中也会出现;(ii)更大的模型和更小的批次规模往往会产生更强的IML效应。我们还通过探测实验研究了IML如何改变模型在参数中存储知识的方式。最后,我们思考了研究结果对未来AI系统的能力、风险及可控性的潜在启示。我们的代码可在https://github.com/krasheninnikov/internalization获取。

英文摘要:

We demonstrate that LLMs may learn indicators of document usefulness and modulate their updates accordingly. We introduce random strings ("tags") as indicators of usefulness in a synthetic fine-tuning dataset. Fine-tuning on this dataset leads to implicit meta-learning (IML): in further fine-tuning, the model updates to make more use of text that is tagged as useful. We perform a thorough empirical investigation of this phenomenon, finding (among other things) that (i) it occurs in both pretrained LLMs and those trained from scratch, as well as on a vision task, and (ii) larger models and smaller batch sizes tend to give more IML. We also use probing to examine how IML changes the way models store knowledge in their parameters. Finally, we reflect on what our results might imply about capabilities, risks, and controllability of future AI systems. Our code can be found at https://github.com/krasheninnikov/internalization.

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。

↑