Tag: Accuracy

Testing Google's AI engine on annuity and life insurance knowledge showed that while it can be accurate, it can also offer incorrect or incomplete answers. This raises concerns about relying on the technology for educational purposes due to lack of transparency on sources. Despite this, 50% of respondents in a recent survey plan to test the models.

OpenAI proposes a new strategy to combat AI models' tendency to produce false or fabricated information called process supervision. Instead of rewarding only the final correct conclusion, AI models are rewarded for each correct reasoning step, improving accuracy in multi-step domains. Their approach could lead to more human-like thinking and better explainable AI.

OpenAI's ChatGPT chatbot failed a US urologist exam with less than 30% accuracy, says a new study. It also spread medical misinformation through errors. The chatbot struggles with clinical medicine questions that require multiple facts, situations, and outcomes.

OpenAI's renowned ChatGPT chatbot recently failed a urologist exam in the US, scoring less than 30% accuracy in the Self-Assessment Study Program for Urology (SASP). Despite excelling in recalling facts, ChatGPT struggled with questions requiring multiple overlapping facts, situations, and outcomes, leading researchers to investigate the limitations of large language models across various disciplines. Find out more about this study in the Urology Practice journal.

ChatGPT and other chatbots could be key players in preventing suicide and helping people quit smoking. Recent research found these AI programs offer effective advice and support for public health issues like addiction, violence, and mental health. While accurate in 91% of responses, only 22% of the time did ChatGPT direct users to appropriate resources.