Is Persona Generalization Predictable?
arXiv cs.CL · 10 Nov 2026, 04:51 AEST
Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others. A surprising 'on-the-line' phenomenon in persona generalization reveals a shared probit-linear relationship across model families and sizes.
Key facts
- 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.
- 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.
- This 'on-the-line' phenomenon unifies unexpected, isolated observations of persona generalization within a quantitative framework for predicting and controlling behavioral transfer.
Read aloud
Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others. A surprising 'on-the-line' phenomenon in persona generalization reveals a shared probit-linear relationship across model families and sizes. 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. 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. This 'on-the-line' phenomenon unifies unexpected, isolated observations of persona generalization within a quantitative framework for predicting and controlling behavioral transfer.
Sources in this cluster
- Is Persona Generalization Predictable?
arXiv cs.CL · 10 Nov 2026, 04:51 AEST