Two Minute Papers on further uses of ChatGPT

the latest Two Minute Papers video (entitled ‘OpenAI’s ChatGPT Took An IQ Test!’, though it is actually mostly not about that) concerns further uses of ChatGPT. i learned from it and will implement today that there are now browser extensions for ChatGPT, which will, i hope, be very useful. also very interesting to see more examples of how ChatGPT can be used to create visual images, including by writing code. i really want something that will create a wide range of graphs and also offer me different styles.

karoly showed a warning example of how ChatGPT fails with basic arithmetic. i have found this to be one of its biggest downfalls – as i mention in my comment below, don’t use it as a calculator. how on earth it manages to create working code (which it often can), and yet be innumerate, i do not fully understand.

here is the comment i posted on the video:

i am now using ChatGPT all the time for research (along with Chatsonic, which is online and has real-time access to the web and google), but you do have to check what it says. some times it gets things totally wrong. yesterday, ChatGPT told me that ‘in Sweden, nuclear power is no longer in use’. flat wrong: it has three power plants and six reactors (as far as i can tell from the web!). ChatGPT was fooled by an earlier intention by sweden to shut down all its reactors, but it (sweden) subsequently reversed this decision and ChatGPT just based its answer on the original decision. when i told ChatGPT that it was wrong, it then gave the correct answer. lesson: at least for now, it is important to cross-check everything, and if you don’t know anything about an area, at least try to read something authorative about it first. and yes, though ChatGPT is supremely literate, it is banally innumerate – don’t trust it with maths!