Overthinking the Truth: Understanding how Language Models Process False Demonstrations