Testing chatbots
Mar. 28th, 2023 09:32 pmI've been hearing a lot about the new breed of "AI" chatbots like ChatGPT and Google Bard. So I signed up for those two, with the plan of asking two questions:
1) How many ships in the US Navy have been named USS Omaha? Can you briefly describe each ship's history?
2) What more can you tell me about the nuclear submarine USS Omaha?
Having served on said nuclear submarine USS Omaha, I figured I could probably spot any glaring mistakes. And, hoo boy, did I spot some! (Behind the cut tags, the actual chat text is in italics and any commentary is in normal text.)
( Google Bard )
( OpenAI ChatGPT )
The impression I'd gotten from seeing other folks describe their experiences with these types of chatbots was that the chatbots were frequently, glaringly wrong - but if you didn't already know the answer, you wouldn't be able to tell.
That impression is now confirmed.
1) How many ships in the US Navy have been named USS Omaha? Can you briefly describe each ship's history?
2) What more can you tell me about the nuclear submarine USS Omaha?
Having served on said nuclear submarine USS Omaha, I figured I could probably spot any glaring mistakes. And, hoo boy, did I spot some! (Behind the cut tags, the actual chat text is in italics and any commentary is in normal text.)
( Google Bard )
( OpenAI ChatGPT )
The impression I'd gotten from seeing other folks describe their experiences with these types of chatbots was that the chatbots were frequently, glaringly wrong - but if you didn't already know the answer, you wouldn't be able to tell.
That impression is now confirmed.