Contextual Contamination: The Silent Drift of Large Language Models via Stored Conversation Data

“Would you be comfortable if I used your contextual contamination setup as inspiration for a small DESi benchmark?“ That sounds interesting if you have questions on the dataset or my set up let me know. Would you mind sharing the process and results? I think it would be really helpful.