#chatbot-arena

AI Models Show Narcissistic Bias in Automated Evaluations

Large language models favor their own outputs in 58% of comparisons, raising concerns about the reliability of automated evaluation platforms like Chatbot Arena.