Papers

研究:通用训练进展可替代被过滤的特定数据,能力可低成本重获

General progress substitutes for specific training data in language models

Why we picked it

做数据过滤或安全训练的团队该看看这篇:论文用国际象棋、Othello 和 7B 模型实测证明,扣掉的技能数据会被模型的通用进步补上,重训时只需很小数据量就能恢复。做模型安全的、管数据集的、想评估过滤策略有效性的都值得对照自己的做法。

arXiv 论文(编号 2611.24844)测试了在训练数据中剔除特定技能数据能否真正阻止模型习得该技能。团队在国际象棋、Othello、分子字符串、括号语言和网页文本等任务上扣留了单个技能的训练信号,例如一个从未被显式训练应对将军的棋类模型,对合法应对着法的排序仍接近被训练过的模型。

Body

结果显示模型的通用进展会替代被扣留的数据:在扣留后重新训练时,模型只需从零训练所需数据量的一小部分即可恢复该技能,且模型越强所需比例越小。一个 70 亿参数模型中被剔除的编程语言数据也以同样方式被重新习得,而一条难度相当的对照规则则学习得慢得多。

更大模型中某技能被延迟的时间可以由较小模型预测出来。对开发者而言,数据过滤只是延迟能力出现,而非根除,且随着模型整体进步,重新习得所需的数据越来越少,这对依赖数据剔除做安全控制的实践是一个量化提醒。

Original · arXiv cs.AI

General progress substitutes for specific training data in language models

Language models acquire their capabilities from training data, and developers withhold data during training to keep models from learning capabilities that could be misused. How long this delays a capability, and how much data is needed to acquire it later, is unknown. Here we show that the general progress a model makes on the rest of its training data substitutes for the data that was withheld. We withheld the training signal for individual skills in models of chess, Othello, molecular strings, bracket languages and web text. The delay of a withheld skill in larger models was predicted from smaller ones. A chess model never explicitly trained to respond to check ranked the legal responses almost as well as a model that was. When models were then trained on the withheld data, they acquired the skill with a small fraction of the data needed from initialization, and this fraction decreased as the models progressed. A control rule of equal difficulty was learned much more slowly. A programming language withheld from the training text of a 7-billion-parameter model was regained in the same way. Data filtering therefore delays a capability, and the data required to regain it decreases as models improve.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include model security, language models, arXiv.

Source attribution: curated from arXiv cs.AI (@Mark Marosi, Peter Antal).