arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.11474cs.CLcs.AI
收藏

上下文内对齐能走多远?探索上下文内对齐的现状

How Far Can In-Context Alignment Go? Exploring the State of In-Context Alignment

School of Computer Science and Technology, Beijing Institute of Technology(北京理工大学计算机科学与技术学院);Beijing Institute of Technology Southeast Academy of Information Technology(北京理工大学东南信息技术研究院)

Heyan Huang, Yinghao Li, Huashan Sun, Yu Bai, Yang Gao

AI总结:

本文通过拆解 ICA 上下文成分并开展消融与零样本评测,发现示例最关键,且 ICA 在知识和工具任务上优于参数微调,但在多轮对话与指令遵循上仍有限。

摘要:

近期研究表明,In-Context Learning(ICL,上下文内学习)通过使用特定示例,可以使 Large Language Models(LLM,大语言模型)与人类偏好对齐,这被称为 In-Context Alignment(ICA,上下文内对齐),表明模型无需调整参数即可理解人类指令。然而,对 ICA 的机制和适用性的探索仍然有限。本文首先将 ICA 中使用的上下文文本分为三类:格式、系统提示和示例。通过消融实验,我们研究了各部分在使 ICA 有效发挥作用方面的效果。随后,我们考察这些部分的变化如何影响模型的对齐性能。研究发现表明,示例部分对于增强模型的对齐能力至关重要,示例的变化会显著影响对齐性能。我们还全面评估了 ICA 在各类对齐任务中的零样本能力。结果表明,与参数微调方法相比,ICA 在基于知识的任务和工具使用任务中表现出更优性能。但它在多轮对话和指令遵循等方面仍表现出一定局限。

英文摘要:

Recent studies have demonstrated that In-Context Learning (ICL), through the use of specific demonstrations, can align Large Language Models (LLMs) with human preferences known as In-Context Alignment (ICA), indicating that models can comprehend human instructions without requiring parameter adjustments. However, the exploration of the mechanism and applicability of ICA remains limited. In this paper, we begin by dividing the context text used in ICA into three categories: format, system prompt, and example. Through ablation experiments, we investigate the effectiveness of each part in enabling ICA to function effectively. We then examine how variants in these parts impact the model's alignment performance. Our findings indicate that the example part is crucial for enhancing the model's alignment capabilities, with changes in examples significantly affecting alignment performance. We also conduct a comprehensive evaluation of ICA's zero-shot capabilities in various alignment tasks. The results indicate that compared to parameter fine-tuning methods, ICA demonstrates superior performance in knowledge-based tasks and tool-use tasks. However, it still exhibits certain limitations in areas such as multi-turn dialogues and instruction following.

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。

↑