How Useful Is Cross-Domain Generalization for Training LLM Monitors?

📅 2026-05-12
📈 Citations: 0
Influential: 0
📄 PDF

career value

179K/year
🤖 AI Summary
This work addresses the limited generalization of prompt-based large language model (LLM) classifiers in data-scarce scenarios, where insufficient fine-tuning often hinders performance. The authors propose a multi-task prompt fine-tuning approach that designs task-specific prompts while integrating general instruction tuning, substantially improving classification accuracy on unseen domains and novel prompts. Notably, they find that supervised classification training without explicit reasoning capabilities can effectively generalize to reasoning-intensive tasks such as summarization. To mitigate performance degradation caused by prompt variations, a hybrid training strategy is introduced. Experimental results demonstrate strong performance on related unseen tasks, highlighting the potential of classification-oriented training for building versatile, general-purpose monitoring systems.
📝 Abstract
Using prompted language models as classifiers enables classification in domains with limited training data, but misses some of the robustness and performance benefits that fine-tuning can bring. We study whether training on multiple classification tasks, each with its own prompt, improves performance on new domains with new classification prompts. We show that such training partially generalizes to adjacent domains, improving classification performance on tasks that are unseen during training. However, we identify specific edge cases where the fine-tuned models fail to follow prompts, such as when the classification prompt changes completely while the data domain remains the same as during training. We show that classification training can be mixed with general instruction following training, and that (when done well) such training keeps the benefits of classification training and mitigates its generalization failures. Surprisingly, we see that this no-thinking supervised classification training can generalize to with-thinking classification and summarization, suggesting that no-thinking classification training might be instrumentally useful in building other kinds of classifiers and monitoring systems.
Problem

Research questions and friction points this paper is trying to address.

cross-domain generalization
LLM monitors
prompt-based classification
fine-tuning
instruction following
Innovation

Methods, ideas, or system contributions that make the work stand out.

cross-domain generalization
LLM monitors
prompt robustness
mixed training
instruction tuning
🔎 Similar Papers
No similar papers found.