In-context Interference in Chat-based Large Language Models