Towards Self-Robust LLMs: Intrinsic Prompt Noise Resistance via CoIPO

Researchers propose CoIPO (Contrastive Learning-based Inverse Direct Preference Optimization), a new method to improve Large Language Model robustness against noisy or imperfect user prompts. The approach enhances LLMs' intrinsic ability to handle prompt variations without relying on external preprocessing tools, showing significant accuracy improvements on benchmark tests.

• →CoIPO method addresses LLM sensitivity to prompt variations by training models to handle noisy inputs internally rather than using external preprocessing.
• →The approach uses contrastive learning to minimize discrepancies between clean and noisy prompt responses using mutual information theory.
• →Researchers created NoisyPromptBench benchmark and augmented FLAN dataset with paired clean/noisy prompts for training and evaluation.
• →Experimental results show significant accuracy improvements over current state-of-the-art approaches for prompt robustness.
• →All source code, datasets, and benchmarks have been open-sourced for community use and further research.