> Gesendet: Donnerstag, 16. März 2023 um 14:13 Uhr > Von: "andy pugh" <bodge...@gmail.com> > An: "EMC developers" <emc-developers@lists.sourceforge.net> > Betreff: Re: [Emc-developers] For 2.10: Suggesting chatGPT checks of our > documentation > > On Thu, 16 Mar 2023 at 12:10, Steffen Möller <steffen_moel...@gmx.de> wrote: > > > Hello, > > > > I am admittedly still impressed of the linguistic skills of chatGPT > > > > It's worth considering what ChatGPT does, which is to a large extent just > choosing the most likely word to continue the sentence. > In many ways this is just a generic critique that could be relevant to > almost anything, with some keywords swapped.
So when our chatty friend encounters a text that is not continuous wrt the sentences it already knows, like the references to the jitter and the descriptions of limitations, then it will point out that this needs some form of deeper embedding the text to cater for a fluent read. This is all I want, really, for now that is. I also expect that ChatGPT points me to redundancies or that it proposes some reordering, but I have not seen this, yet. There is a side-thought on this. I want our documentation "understood" (are those quotes still needed?) by a linguistically skilled robot. That robot could then help our new users to set their mill up and (e.g. by integrating the forum and I also hope for more YouTube transcripts) also help with error recovery. > Interesting examples: > https://economistwritingeveryday.com/2023/01/21/chatgpt-cites-economics-papers-that-do-not-exist/ Yes, it is a complete utter failure on provenance, i.e. the introspection on where a particular information was found/derived from. But that will come. Best, Steffen _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers