Ask a chatbot for sayings from the Gospel of Thomas and you will get a generous list. Most of them will be real. One or two will be lovely, serene, perfectly on theme, and found nowhere in the Gospel of Thomas. The machine does not know it is lying, which is exactly what makes it dangerous to a student of ancient texts.
I use AI to study these texts almost every day, and I am not giving that up. It is the most patient reading partner I have ever had, and the fastest. But the partnership only works under rules, and the rules took some burned fingers to learn. Here is the whole workflow, including the part where the partner gets caught inventing scripture.
Why These Texts Need a Partner
The Nag Hammadi library is genuinely hard to read alone. The texts arrive full of unexplained jargon, aeons and archons and pleroma, and they assume a mythology no modern reader grew up with. Pages are missing, sentences break off mid-thought, and the Gospel of Thomas is a bare list of sayings with no story to orient you. The scholarship that explains all this mostly lives in expensive academic volumes.
A tireless assistant that has read those volumes changes the experience completely. Questions that once required a seminary library now get answered at midnight in your kitchen. That access is real, and I refuse to be cynical about it.
What the Machine Is Actually Good At
Comparing translations is the killer feature. Bentley Layton renders gnosis as acquaintance where most translators write knowledge, and asking why that choice matters opens up the whole heart of the tradition. An AI can lay the versions side by side and explain what each translator was reaching for. Doing the same thing with physical books takes an afternoon and a large desk.
Cross-referencing comes second. Ask where else the bridal chamber appears across the codices, or which sayings in Thomas have parallels in the canonical gospels, and you get in seconds what once took a concordance and a week. Themes that run quietly through separate texts, the sleeping soul, the forgotten garment, the seed of light, suddenly become visible as themes.
And it tolerates ignorant questions with perfect grace. You can admit you have forgotten what an aeon is for the fourth time. No seminar room on earth offers that.
The One Rule That Matters
Never publish, repeat, or believe a quote you have not seen with your own eyes in a printed translation. Not because the machine is malicious, but because it is fluent, and fluency fills every gap with plausibility. When its memory of a text runs out, it does not stop talking. It rounds the corner smoothly and keeps going, and the seams do not show.
My ground truth is Layton’s The Gnostic Scriptures, and every quotation on this site gets checked against it before publication. The AI may point, but the book decides. This single habit catches everything that matters.
Rules That Catch the Fakes
A few smaller rules back up the big one, and they transfer to any field where a machine becomes your research assistant.
Demand the address, not just the words. Ask for the saying number, the codex, the page. Fabricated quotes usually live at fabricated addresses, and a wrong address is faster to catch than wrong wording.
Distrust the perfect quote most. If a line fits your thesis like a tailored suit, check that one first. The model is trying to please you, and scripture on demand is the most pleasing product there is.
Watch for smoothness. Real ancient texts have edges. They break off mid-sentence, contradict themselves, and say bewildering things about lions. A quote with no strangeness in it deserves a second look, because the fakes are always smooth.
Make it argue against you. Ask what a skeptical scholar would say about your reading. The same fluency that invents quotes is surprisingly good at inventing objections, and objections are safe to use without checking.
Keep the book on the desk. Whatever your field’s version of Layton is, own it in print or PDF. The final step of every session is a page number, and the machine never gets the last word.
A Session From Last Week
Concrete beats abstract, so here is a real one. While drafting the algorithmic archons post, I wanted the saying where Jesus finds the world drunk. The AI produced it instantly, in wording I had seen quoted a hundred times online, ending with the line that when the people shake off their wine, they will repent.
Layton’s page told a slightly different story. His translation ends with they “will have a change of heart,” and it opens with “I stood at rest in the midst of the world” rather than the more familiar “I took my place.” Small differences, but not nothing. A change of heart lands as diagnosis where repent lands as scolding, and standing at rest quietly echoes the repose language that runs through the whole tradition.
That is the partnership working as designed. The machine found the saying in seconds, and even its secondhand wording was close enough to locate the real thing. The book supplied the exactness, and the exactness turned out to carry the meaning.
What Study Cannot Do
All of this stays on one side of a line the tradition itself draws. Study piles up information, and the texts insist that what saves is acquaintance, the kind of knowing that meets rather than accumulates. No model can do that meeting for me, and a mind that has read everything while being acquainted with nothing is the tradition’s own cautionary figure. The study partner is a partner for study, and study was always the smaller half.
So the last step of my workflow has nothing to do with verification. Close the laptop, keep the one saying that stung, and sit with it in silence for a while. The machine can hand you the text. It cannot be the one the text happens to.
The Fakes Are Always Smooth
Back to that invented saying from the opening. The counterfeits are consistently serene, balanced, and a little bland, like stock photography of wisdom. The real sayings have splinters. Jesus calls the world drunk, blesses lions, praises passers-by, and says almost nothing the way you would expect.
Strangeness, it turns out, is the fingerprint of the genuine. Any machine can produce something that sounds like scripture, but no machine produces the line that has been bothering people for eighteen centuries. When a saying refuses to flatter you and still will not let go, you are probably holding the real thing. Hold on.
Quotations from the Gospel of Thomas are from Bentley Layton’s translation in The Gnostic Scriptures (Anchor Yale Bible Reference Library).
