ChatGPT, ce matin, après 10 minutes de conversation avec moi 👇.
(Et pour joindre l'agréable à l'utile : lire cet article « L'intelligence artificielle n'existe (toujours) pas ».)
#chatgpt #chatgpt_v4 #chatgptfails #humour #humor #intelligenceartifielle #artificialiintelligence #ai #ia

Tested ChatGPT with summarising an old abstract … wow its really really not very good is it! On the surface it looks quite impressive but almost all the detail is lost - very good for superficial 'science’ but extremely worrying if it gains more widestream acceptance. Dropped all the actual data from the results section also. Avoid! #AI #chatgptfails

Tried #chatGPT and documenting here my first conversation #ChatGPTfails

As models with stronger safeguards keep being subverted by people determined to find workarounds and make the model say bad/funny things, the conclusion I'm drifting toward is not that models are dangerous but that this order of animals can't be trusted with anything. #ChatGPTFails

The large number of #ChatGPTfails wouldn't be worth mentioning if it weren't for the large number of people who find an impressive *non*-failure and draw ill-advised conclusions about the model's suitability as a knowledge source. Here's me piling on with the first thing I tested the new model on. I suspected that since this puzzle is always (?) expressed in terms of the North Pole (so a twist concerning bear colour can be added), ChatGPT would fail to abstract to a South Pole setting.