My AI LXX and Other Texts Translation Adventures

I’ve been busy translating Genesis from the LXX. My original plan was to translate the entire Rahlfs 2006 updated edition. I started in December 2024, and I’m just now getting to the finishing touches—four more chapters of translation notes, and I’ll be done. But I’m 60, and at this rate, Jesus will either have returned, or I’ll be dead before I get that far! But I will go as far as I can.

One reason it’s taken so long is that I wasted who knows how many hours trying to find an AI solution to assign Strong’s numbers to words in the LXX that already exist in Strong’s. I tried GPT. I tried Grok. I tried every method I could think of. I finally gave up sometime in January. AI kept assigning numbers to words not in Strong’s and missing words that were. Nothing worked.

So I gave up and decided to create detailed translation notes and put them into a commentary module instead. I plan to put both the translation and the translation notes into eSword and give it to BibleSupport.com. So, I am creating the necessary modules to do that.

And let me tell you, those notes are what have taken up the most time. The translation itself only took about six weeks. But organizing the notes into something coherent and figuring out a consistent format? That’s been a real challenge.

Let me talk about this adventure. I’ve learned a lot. Translating has opened my eyes to God and His world in a way nothing else ever has. It’s been amazing—indescribable, really. He and Jesus have become my reality in a way I had never experienced before. Truly understanding the Gospel starts in Genesis. Seeing how God manages everything down to the smallest details—it’s all there. Genesis reveals that His power exceeds every known limit—and even the ones we don’t know. I could go on and on.

This journey propelled me to translate some New Testament books as well. I’ve started with Ephesians and Philippians. Wow. What an eye-opener. It’s simply amazing.

None of this would have been possible without AI. I do have a working knowledge of Greek—basic stuff like moods, verbs, nouns, rough breathing marks, the alphabet. I can spell words (though I can’t pronounce them), and I understand diphthongs, vowels, and things like that. I use the polytonic Greek keyboard that comes pre-installed on both PC and Mac, which lets you type in Koine Greek with all the breathing marks and accents. I definitely couldn’t write Greek by hand! Without that keyboard Microsoft includes with their OS, I’d be lost. I’m not a Greek scholar—probably not even at a first-year student level. All of this is self-taught.

So I use AI with that basic Greek knowledge and know when to ask questions to clarify things. My head has been absorbed with the Bible for a long time, and I’ve read it from cover to cover more times than I can count. I’m not saying this to boast—only to say that without at least some of this background, using AI for Bible translation is a fool’s errand. You cannot blindly trust AI. It will hallucinate, and if you can’t spot those hallucinations, you’ll end up with a mess. You have to explore everything it gives you, ask the right questions, dig deep. If you accept what it spits out without question, you’ll get blind man results. That’s all I’m saying. And you must compare everything with established translations—I use four different ones for that very reason.

Let me give you a prime example of what I mean by hallucinations and “blind man results.” It happened to me.

This translation work has gripped me hard. I’m at it 10, 12, sometimes 16 hours a day, 7 days a week. The Book of Jubilees has always fascinated me, but I’ve never been happy with the current English translations. So, I thought, “Why not do it myself?” Jubilees is based on the Ethiopic Ge’ez text, which is a language I know absolutely nothing about. However, after completing my LXX Genesis translation, I felt confident in my process and ability to ask the right questions. Famous last words!

I’m now 25 chapters deep into translating Jubilees. It’s been slow going, because my translation would often differ from other versions—sometimes every other verse. So I’d get the Ge’ez breakdown for each verse. Sometimes I’d find the mistake and bring my translation in line with the others. But most of the time, I wouldn’t.

So I had GPT compare my translation against the other version and the Ge’ez text. It would generate a line-by-line comparison, put it all in a table, and occasionally agree with the other translation—sometimes revising mine and explaining why. MOST of the time, it would say the other version took interpretive liberties, while mine stuck closer to the literal Ge’ez. I accepted that and began writing notes explaining where the other translation strayed—sometimes outright ignoring the Ge’ez and adding in whatever they wanted.

Then I came to a verse where our translation made absolutely no sense. No matter how I reframed the question, GPT just wasn’t coming up with anything better. Now, Genesis and Jubilees have a lot in common—it’s often called “Little Genesis.” It has a stronger covenant focus and gives more detailed family info than Genesis. But they generally agree on the big stuff. So I knew something was off. GPT was willing to bend the text in the direction I thought it should go—but I have a strict “let the text speak for itself” rule. I won’t let GPT modify anything based on my assumptions.

Still, this verse was clearly wrong. So I tried Grok. Boom—Grok produced the same translation as the version I had been comparing to all along. Meanwhile, GPT was out in left field—or on the moon.

So I spot-checked a few other verses that had felt off. I hadn’t pushed on them before because I thought the other translations were taking liberties or skipping things. But nothing was theologically problematic—until now. And again, Grok’s translations matched the standard one closely.

So I started questioning Grok. That’s when I learned Ge’ez is a minor language used mostly in the Ethiopic church, with few manuscripts and limited scholarly resources. AI in general, including Grok, isn’t fluent in Ge’ez. Grok explained that GPT was hallucinating—essentially guessing—and doing it in a way that seemed logical and convincing. That blew my mind… and cost me about 100 hours.

That’s why I say: you should probably know something about the language and the Bible before trying to use AI for translation. For me, because I know Genesis so well, it helps keep me grounded. I’d actually argue that knowing the Bible well—from cover to cover, in multiple English translations—is more helpful than just knowing the original language. This whole experience has been wild. And even though my method for arriving at a final translation was solid, GPT was hallucinating—like it was dropping LSD—and making everything sound legit. That’s why you have to be extremely careful when using AI to translate the Bible, Second Temple period texts, or early Christian writings.

Yeah… this has been a real adventure, that’s for sure!