I’m having good interchanges with google’s gemini, while I struggle with Isocrates 4. Gemini seems to have good knowledge not only of Greek grammar, but also of the political-historical environment in which this speech was made. It is infinitely patient with my followup questions about grammar & syntax.
We had a disagreement about the conventional citation system for Isocrates’s orations, but when I referenced the Perseus citations, Gemini adjusted its references to Perseus.
Gemini readily answers questions about its own data handling, and abilities.
It makes me wonder if there are “taboo” questions, e.g. can AI do the work of billionaires, but at much less expense?
Good point cmw, but gemini (google’s ai) does accept correction. Moreover, a reference to the session between the user and gemini can saved, so that the user can resume the discussion later. I am an AI skeptic myself, but I’m not well equipped to critique gemini’s answers, because I don’t know enough–after all I went there because I was stumped.
I believe that in the subjects I know best I can responsibly critique gemini, but I probably would not go to gemini for help in those subjects; instead I’d do a web search, search google scholar, and the like.
In AG I use use translations when baffled. I can usually match the words in the text with the words in the English translation, and often with this match I can guess the grammar topics I need to review. But, when well and truly stumped I may not find the words I need for searching the indices.
Try to discuss metrical analysis with it, and you will see some limitations. It will propose things that are totally wrong with extremely high confidence. When you point it out, it will correct itself, but most often to something equally wrong.
Accepting correction isn’t the same as offering unprompted corrections. There have been times when I followed up with my class, client, colleague, or even just friend, apologizing for my mistake after further review. No one called me out. No one caught the error. I just read something, realized my mistake, and reached out to correct it. That’s what I mean.
I can use AI to help as a sounding board, to quickly spit out some information, or write some code for me, but anything it gives must be verified twice, because it doesn’t really know what it’s saying.
Yep, figured this out early on. It couldn’t even count syllables. It’s just not in its training data (or wasn’t last time I bothered with those types of questions).
I want to show the kind of questions I ask google ai.
To see how google ai answers ancient-Greek queries, cut and paste a sentence into the google search box, precede it with something like “need a grammatical analysis”. Or put in a phrase and ask for a translation.
The reply gives spaces for followup questions. You can ask questions like “how do we know that [some word] is not a participle?” Google ai doesn’t care how stupid my questions are.
Or, “how did the ancients, before punctuation, know that this sentence is a question?”