What percentage of ChatGPT's output is wrong? There's no totally objective way to test that, but I gave it a shot. Having never used ChatGPT before, I sat down on the couch and wrote out 16 questions designed to test the system in specific ways. There were a few pleasant surprises, but the answers were seriously inaccurate most of the time, and for more than a third it just made up total bullshit.
https://ben-crowell.bitbucket.io/testing_chatgpt/
#AI #AIethics #chatgpt
Factually wrong might not be useless. Let me invite you to have a look at why it is wrong, and why it may matter less then we think it would (spoiler: all facts need checking, and at least it is not malicious). See here: https://sentientsyllabus.substack.com/p/chatgpts-achilles-heel
