I’m working through the letters of Procopius of Gaza. He was a sophist living in the early 6th century, after the end of the Origenist disputes, and before the rise of Islam. Only three of the letters are addressed to priests, and the tone is secular. But he lived in a period when the traditional Roman upper class was starting to be replaced by the ecclesiastical dignitaries, themselves rich and powerful and full of patronage. In other words, he lived at the changeover period between the Roman and Byzantine periods.
Initially I was rather charmed with the letters. I got hold of the Italian translation, and ran the lot through DeepSeek, the Chinese AI, so that I could have a read. I’ve tended to find that DeepSeek produces less “wild” results than ChatGPT, so I use it. The translations are readable enough. I thought that others might find them interesting also.
This has led me to start to run each letter in turn in Italian through Google Translate, and compare with the DeepSeek output. In Latin I have often found that Google Translate is closer to the Latin. It’s also based on a neural net technology, rather than AI, so the comparison ought to reveal hallucinations in the DeepSeek output. So far, after 57 letters, I have discovered no hallucinations.
Interestingly, Google Translate is often producing more focused language than DeepSeek. The latter can waffle around something which could be expressed more concisely and clearly. Some of the AI stuff makes your eyes close. You read the words, but they convey nothing to the mind.
I then got hold of the TLG Greek text of letter 1, and ran that through DeepSeek and Google Translate. The latter was futile – Google Translate doesn’t support ancient Greek. The former produced an interesting output, and I compared it to the evolving translation for letter 1. It revealed that the Greek is quite a bit more concise than the Italian. In a couple of places I preferred the output from Greek.
Originally my intention was simply to run the Italian translation through DeepSeek so that I could read them myself. I’d add a few notes; and then throw the thing over the wall, so that others could easily do the same. How else, after all, would any normal person be able to engage with Procopius’ letters? It’s not a translation as such; but it makes life easier for the researcher. And this, I thought, might take a day to do.
Once you start doing this, and comparing other outputs, the timescale stretches out. It’s starting to get slow, and cumbersome.
I’m also getting rather fed up with Procopius. All the letters so far seem to be the same letter – the author whining to some correspondent that he hasn’t had a letter from him, suggesting sometimes that perhaps the latter is now too rich and important to reply, adding maybe a classical allusion, and so on. I’ve worked through 57 of these, and I am getting rather jaded. The historical content is nil. Are these letters really just rhetorical exercises, saying “look at how nicely I can write Greek!”? Maybe.
So I may cut this short, and do less from here on. We’ll see.
I always take the view that whatever I do is a step forward from whatever we had before, because there was nothing there before. I also consider that I have no obligation to do more than I feel like doing. There are people out there who are paid to do this stuff, after all!
So I won’t hesitate to shirk if I have to. Sorry Procopius!