More #AI testing. Repeating my tests from a couple of months back of #ChatGPT (ChatPDF to be precise) on #astronomy papers, but this time using #Bing, which uses the more advanced #GPT4.

I have to say this is quite the improvement. ChatPDF invents stuff at the drop of a hat. Bing does still make stuff up, but much less frequently. In the papers I tested, it got all the major scientific points right. Even when I thought it'd got something wrong, further questioning revealed it was right but just not expressing itself in an ideal way. It does not do well with figures in PDF files, however, even though it thinks it can, and using it in "balanced" mode is truly excruciating.

My conclusion from the ChatPDF tests was that it was tech demo stage, not actually useful. Can't say the same for Bing - this at the "I plan to actually use this myself" stage already. It can't do my work for me, but it can definitely help.

https://llittlephysicists.blogspot.com/2023/05/bing-versus-me.html