Vectara has created an AI hallucination leaderboard, ranking AI chatbots on their ability to avoid fabricating facts. The leaderboard, updated periodically, currently shows GPT-4 as the most accurate, while Google’s Palm-Chat has a hallucination rate of over 27%. The evaluation model could become a benchmark for those using large language models for non-creative tasks. Elon Musk’s recently launched chatbot, Grok, is expected to be measured soon.
