Saturday, January 28, 2023

Large Language Models like ChatGPT say The Darnedest Things

As the threats and opportunities of ChatGPT are being discussed, here is a blog post on the ACM website, highlighting types of mistake (e.g. in arithmetic) made by ChatGTP https://cacm.acm.org/blogs/blog-cacm/268575-large-language-models-like-chatgpt-say-the-darnedest-things/fulltext. In this post they mention a database they are compiling (ChatGPT/LLM Errors Tracker) of the prompts and erroneous responses that people have got from ChatGPT, which can be found at https://researchrabbit.typeform.com/llmerrors?typeform-source=garymarcus.substack.com

Photo by Midjourney/Sheila Webber, created with the prompt mistakes errors sadness computers chatgpt

No comments: