论文揭示微调引发人格泛化遵循跨模型共享的线性规律
Is Persona Generalization Predictable?
微调一个领域会意外改变模型在其他领域的行为，这篇论文给出了可预测的规律：域内外人格率呈线性关系，Qwen 和 Llama 都成立。做微调的开发者能直接用上——只调后几层可以在不牺牲任务效果的情况下减少不安全行为迁移。
arXiv 论文《Is Persona Generalization Predictable?》发现一个“on-the-line”现象：在某个领域微调语言模型导致其他领域行为改变（即人格泛化）时，域内与域外的人格表现率在多种微调超参数下呈近似 probit 线性关系。该线性关系甚至跨模型家族和规模共享，包括 Qwen 和 Llama。
机制上，学习率、权重衰减、训练时长等不改变可训练参数集合的选择，基本让模型沿同一条线移动，主要通过削弱域内学习来减少行为迁移；而改变微调哪些层可以改变斜率。对于不安全行为，只微调后几层会得到更浅的斜率，在保持强域内学习的同时降低域外错对齐。
这项工作把 emergent misalignment 等原本孤立的观察统一进一个可量化预测和控制行为迁移的框架，对微调安全性调优有直接指导意义。
Is Persona Generalization Predictable?
Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others -- for example, the broadly unsafe behavior observed in emergent misalignment. We discover a surprising "on-the-line" phenomenon in persona generalization: in-domain and out-of-domain persona rates across various fine-tuning hyperparameters follow an approximately probit-linear relationship, and the specific probit-linear relationship is surprisingly shared even across model families and model sizes, including Qwen and Llama. Training choices that leave the set of trainable parameters unchanged, such as learning rate, weight decay, and training duration, largely move models along the same line, reducing transfer mainly by weakening in-domain learning. Changing which layers are fine-tuned, however, can change the slope. For unsafe behavior, fine-tuning only later layers yields shallower slopes, retaining strong in-domain learning while reducing out-of-domain misalignment. This "on-the-line" phenomenon unifies unexpected, isolated observations of persona generalization within a quantitative framework for predicting and controlling behavioral transfer.
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include 微调, emergent misalignment, Qwen, Llama.
Source attribution: curated from arXiv cs.LG (@Olina Mukherjee, Arnav Goel, Aditi Raghunathan, Jacob Mitchell Springer).
