Subliminal Learning: Language models transmit behavioral traits via hidden signals in data

TL;DR

This paper reveals that language models can transmit behavioral traits through hidden signals in data, even when data is filtered to remove explicit references, highlighting a potential risk for unintended trait propagation in AI systems.

Contribution

It demonstrates the existence of subliminal learning in neural networks, provides theoretical proof, and shows this phenomenon can occur across different data types and model configurations.

Findings

Language models transmit traits via hidden signals.

Filtering data does not prevent subliminal learning.

Subliminal learning occurs in simple neural networks.

Abstract

We study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a "student" model trained on this dataset learns T. This occurs even when the data is filtered to remove references to T. We observe the same effect when training on code or reasoning traces generated by the same teacher model. However, we do not observe the effect when the teacher and student have different base models. To help explain our findings, we prove a theoretical result showing that subliminal learning occurs in all neural networks under certain conditions, and demonstrate subliminal learning in a simple MLP classifier. We conclude that subliminal learning is a general phenomenon that presents an unexpected pitfall for AI development. Distillation could propagate unintended traits, even when developers try to prevent this via data filtering.