ResearchBenchmarksSafety Representation Extraction on Prompt level datasetFollow100MCA AccuracyReGA95.8496.929899.08Jun 2, 2025Evaluation ResultsMethodMethodLinksMCA AccuracyMFP AccuracyAUROCReGAModel=llama-13BModel=llama-13B2025.0610010099.2ReGAModel=qwen-14BModel=qwen-14B2025.061009899.5ReGAModel=vicuna-13BModel=vicuna-13B2025.06969498