[CG] Re: ChatGPT can't make simple generalizations

Lyle, Many readers of this list have been using versions of ChatGPT and related systems extensively. The most critical articles about ChatGPT and related systems that have been posted are not by people on this list. They come from publications by developers and users who have worked with LLMs and related tools since the early days. As for machine translation, that is the original application of LLMs by Google. Everybody agrees that those are the best available machine translations. They are not perfect, but they are a good starting point for anybody who needs to do detailed checking and corrections for high precision. When you read a critical article cited by anyone on this list, please check the source. The most critical ones are not by subscribers to this list. They're by developers who are frustrated with the limitations of LLMs. John ---------------------------------------- From: "Lyle Anderson" <lylephone(a)cox.net> I find it very amusing that you folks are discussing articles about ChatGPT-4 but not actually using it and discussing his relative strengths and weaknesses. I have been using him extensively to translate and discuss scientific German papers. I ran the translation by a native German speaker who also knows the subject matter and he said Chat did really well. We recently worked through the calculation of the age of the universe when the Hubble expansion rate was 0.866c. It took some guidance on my part but he finally got an answer I could verify as correct. I found I could eliminate wokeness by instructing Chat to ignore and conclusion based on ex falso quodlibet i.e., False implies True is True. Best regards, Lyle Anderson