Inoculation Adapters: Improved Selective Generalization of Capabilities with Fewer Surprising Backdoors
Inoculation prompting is a selective-generalization technique used against Emergent Misalignment. We introduce inoculation adapters (IA), a family of methods that similarly reduce the optimization pressure to learn undesired traits by strengthening those traits during training. Inoculation adapters are LoRAs that are trained and used in three steps: (1) trained on undesired traits; (2) attached frozen while a separate task adapter is trained on data exhibiting both desired and undesired traits; (3) the IA is discarded at deployment, while only the task adapter is kept. We compare inoculation adapters with four selective-generalization baselines: inoculation prompting, preventative steering, Concept Ablation Fine-Tuning (CAFT), and KL regularization. Across nine setups and five model families, the inoculation adapter family spans a new Pareto frontier of desired trait retention vs. undesired trait suppression, although given wide confidence intervals the magnitude of improvement remains uncertain. Inoculation adapters also avoid two drawbacks of inoculation prompting: they can suppress capabilities and traits that cannot be reliably elicited by a prompt, and they introduce fewer surprising backdoors. However, no IA variant optimizes all objectives perfectly; gains in desired-trait generalization are generally accompanied by weaker suppression of the undesired trait and increased backdoor occurrence.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Inoculation Prompting: Eliciting traits from LLMs during training can suppress them at test-time
Language model finetuning often results in learning undesirable traits in combination with desired ones. To address this, we propose inoculation prompting: modifying finetuning data by prepending a short system-prompt in…
Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment
Large language models are sometimes trained with imperfect oversight signals, leading to undesired behaviors such as reward hacking and sycophancy. Improving oversight quality can be expensive or infeasible, motivating m…
Improving QA Model Performance with Cartographic Inoculation
QA models are faced with complex and open-ended contextual reasoning problems, but can often learn well-performing solution heuristics by exploiting dataset-specific patterns in their training data. These patterns, or "d…
modelProactive Defense: Compound AI for Detecting Persuasion Attacks and Measuring Inoculation Effectiveness
This paper introduces BRIES, a novel compound AI architecture designed to detect and measure the effectiveness of persuasion attacks across information environments. We present a system with specialized agents: a Twister…
Prompt EngineeringCausal InferenceFisherTune: Fisher-Guided Robust Tuning of Vision Foundation Models for Domain Generalized Segmentation
Vision Foundation Models (VFMs) excel in generalization due to large-scale pretraining, but fine-tuning them for Domain Generalized Semantic Segmentation (DGSS) while maintaining this ability remains challenging. Existin…
Semantic SegmentationVariational Inference