Shifting the Gradient: Understanding How Defensive Training Methods Protect Language Model Integrity

TL;DR

This paper compares two defensive training methods for language models, PPS and IP, revealing they operate via different mechanisms and have distinct effects on trait expression and model behavior.

Contribution

It provides the first behavioral and mechanistic comparison of PPS and IP, clarifying how each method defends against trait acquisition in language models.

Findings

PPS shifts activation gradients to attenuate trait expression.

IP's gradient signature is more diffuse and less mechanistically understood.

PPS can reduce pre-existing trait expression, unlike IP.

Abstract

Defensive training methods such as positive preventative steering (PPS) and inoculation prompting (IP) offer surprising results through seemingly similar processes: both add trait-inducing objects to large language models (LLMs) during training, and both defend the LLM against acquiring the trait. The surprising success of these methods comes with the question: how do they work? Are PPS and IP doing the same thing? We provide behavioral and mechanistic comparisons of these two methods using "evilness" as a case-study trait. Our central finding is that PPS and IP achieve their defensive benefits through distinct mechanisms. Behaviorally, we show that neither PPS nor IP operates through a purely associative mechanism; and PPS can both defend against trait acquisition and actively reduce pre-existing expression, whereas IP is ineffective in models that were previously finetuned to express the trait. This behavioral divergence is reflected mechanistically: PPS shifts the activation gradient towards an attenuating direction along the PPS vector axis. When the PPS vector is aligned with a trait-expressing axis, it can reverse the gradient pressure, reducing rather than increasing activation along that axis. In contrast, IP continues to resist a precise mechanistic account. Direct cosine similarity analyses reveal that IP has a characteristically different gradient signature than PPS, and qualitative analyses reveal IP's gradient to be more diffuse. Furthermore, IP reduces the next-token prediction loss on trait-expressing data where PPS need not, consistent with the notion that IP "explains away" the trait-expression in the training data. Taken together, our analyses reveal distinct mechanisms by which each method operates and highlight open questions about IP's mechanistic picture.