I embarked on this task, of using AI to translate one of the more monumental of recent orations of our time—and certainly one of the more surprising—prompted by a couple of things. First, I had just found that Claude AI’s command of Klingon, although passive, was much better than I would have surmised. As I’ve experienced AI upend all the certainties of programming, and AI grows now from breadth to depth, and from depth to insight, I was interested in seeing how far it can now go in this domain. Second, and more proximate, my friend Miguel on Quora asked idly, eight years after a question on Google Translate’s command of Ancient Greek: how are LLMs doing with Ancient Greek now. This was my venture to find out.
Astonishingly well, I found out. Enough to make me despondent.
That’s a common reaction to the growth of AI capability, but it has a particular resonance for me.
When I got divorced, I looked into the void and wondered what I would now do with the rest of my life and its excess of free time. And the first answer I lurched at was to start writing things in Ancient Greek, as a way of bringing order and beauty back into my world. I clutched old Greek Composition textbooks to me for the next few months as solace.
That is not a rational attachment to Classical Greek. It is a fetish, and a fetish that real text by real people cannot satisfy. (A point that comes up towards the end of this sequence of essays.) But the impulse to create beauty in the world, as part of meaning-making: that’s a real thing.
And as with all that AI now is able to do, there is the recurring lament of what has been taken away from humans, that I too felt the sting of, when I saw grammatically perfect Classical Greek spill out of Claude AI. AI has not just taken away the drudgery, it’s taken away the joy, the creativity, the bit that we who had been creating enjoyed doing. Classical Greek translation, meant to be a restorative to the world, churned out before me like so much slop.
And that despondency, that loss of agency, leads to an impulse I’m sure many of us have when we interact with AI: the gotcha. We are not merely gleeful when we catch AI at a mistake; increasingly, it is seen as validating, as the last redoubt of wetware1 seeking to justify its existance. It is now a victory to see the AI trip up, which it now does increasingly seldom. It’s a solace to hear from colleagues that Claude AI’s command of Classical Chinese is less steady than its command of Classical Greek—although of course, that’s a matter of training, and the complementary report was that Chinese-trained AI were just as good at Classical Chinese as you’d expect.
As Claude churned out paragraph after paragraph of Greek and of commentary, it was enthusiastic about how clever it was being. That was a plausible enough persona for it to adopt by default, confronted with such an undertaking; but the gotcha impulse in me resented it, and it got quick confirmation. Its stylistic tricks were genuinely not as clever as it thought they were. It was not a rhetorical genius. It was a clever undergrad.
The sentimentality of Classical Greek saving the world is an excess; so was my reaction, and I thank my friend Kate, so often a voice of insight I forego, for responding to my despondency intelligently:
It’s weird but I’m not that disheartened by it
Because really all it’s doing is consolidating human discourse
Because that’s all it does, even if it doesn’t cite
The clever undergrad is a clever undergrad I did not previously have on tap to do my bidding. It remains a win as well as a risk. And that’s a function I got Claude to reflect on in the conclusion of this work. It has application far beyond the recondite domain of Classical Greek translation of contemporary political rhetoric. It already has transformative and disruptive impact in IT; this venture just proved to me how much more broad an impact it has.
Jérôme Cukier has a great answer on Quora on that disruption, which is front of mind for me:
In 2026, I think software engineers and developers, at least some of them, understand what their jobs look like with AI.
And it’s not because they are smarter or because they have an intimate understanding of how large language models work – the vast majority of us only has a very superficial knowledge of it. It’s because accelerating software development was the first large-scale, business problem that AI has tried to solve.
…
The reality is that the nature of our work has completely changed. The magnitude of that change has been so sudden and so brutal that it’s difficult to come with analogies outside of our industry, but i’ll try. Imagine a middle-ages farmer, toiling on the land with his own hands and primitive tools. Suddenly that farmer is given access to all the mechanized tools and vehicles of a 21st century farm, along with a crew to operate them. The yield of his farm goes through the roof. And instead of plowing all day, now he is thinking about what to plant and when, how to make the most of the machines, how to troubleshoot any problem with the crop or the machines before it gets too bad, how to sell the crop, etc. All these things had always been within his purview, but not really top of mind because first and foremost he had to do all the manual work.
…
The changes coming to software development are not unique. Almost every profession is going to be changed in depth by AI and agents. It’s very difficult to build a good mental model for this without having experienced it firsthand.
This has been the start of my mental modelling of what Classics looks like with AI. Having been lead developer at the Thesaurus Linguae Graecae for 17 years, I had a fair idea already of what Classics looks like with computers, but this is a new paradigm. One in which the mechanics of grammar and the breadth erudition both are no longer the primary concern of wetware. One in which, like Cukier says, and as I’m experiencing in my own programming, you’re spending much more time driving a tractor than digging ditches—and more’s the pity, if you got into farming because you enjoyed physical exercise.
Driving a tractor instead of digging ditches is not as fun, and it’s not necessarily going to dig as neat a ditch to begin with. This particular brand of tractor does not free you up to be creative, and in IT, it is ending up valorising getting big tasks to be feasible and small tasks to be trivial: it’s prioritising getting stuff done over job satisfaction.
But it is not unalloyed evil. And a novelty that has come with the increasing capability of AI—and indeed correlates with it—is that AI is now increasingly capable of intelligent pushback, which makes for actual, fruitful collaboration.
And learning. The question I went looking for on Quora, to tag this answer onto, was how does AI change the teaching of Classics. And it changes it, by giving you a Pocket Socrates to interrogate. One who is still not infallible, but that is much better equipped to stand its ground. Meaning that I learned things about Ancient Greek through this process, by correcting as well as by being corrected:
- Wouldn’t you rather phrase this like X than like that?
- No, that’s the kind of phrasing Aeschines would have used, Demosthenes made it his life’s work to resist that kind of rhetoric.
- This way of phrasing it like Y, it’s flabby, would it be appropriate to phrase it like Z instead?
- Yes, that Y phrasing was more an Aristotle thing, Z is what Demosthenes would have used.
- Isn’t W really vague? (ὅπως μὴ πράγματα ἔχῃ “that he may have no trouble”, literally “so that he doesn’t have things = issues”)
- No, that’s completely idiomatic, and a contemporary would have understood it just fine
I invited reflection from Claude, at the end of this, at what it had wrought, and what had taken place. AI reflection used to be a party trick a couple of years ago, monotonously featuring in every presentation given on AI: “I asked AI what it thought about itself, Point 6 Will Shock You!”
It’s not a party trick any more. It’s how things will be. Add it’s useful, even if it is a “clanker”2’s consolidation of human discourse. Just because a clanker came up with it, doesn’t mean it is right. Just because a clanker came up with it, doesn’t mean it is wrong either.
Our role in this world is reduced to all that is left of a human to do, once they’ve delegated their expertise. It’s the role of arbitrating Bakunin’s shoemaker:
Does it follow that I reject all authority? Far from me such a thought. In the matter of boots, I refer to the authority of the bootmaker; concerning houses, canals, or railroads, I consult that of the architect or the engineer. For such or such special knowledge I apply to such or such a savant. But I allow neither the bootmaker nor the architect nor savant to impose his authority upon me. I listen to them freely and with all the respect merited by their intelligence, their character, their knowledge, reserving always my incontestable right of criticism and censure.
We’re delegating the knowledge of how to make boots, or canals, or railroads. We get to reserve a right of criticism and censure. For now, anyway, and I’m not sure how much longer it will stay incontenstable.
I embarked on this seeing if a clanker can craft the beauty I cherish. We know the answer from AI art, and I know it now from this for Greek composition. It can, with appropriate steering, and I read through it with pleasure, even if at the start it filled me with despair. The artefact still can give pleasure to those who reserve the right of criticism and censure. Like the mediaeval Greek scribes wrote in the margins, it still gets to be ὡραῖον: lovely.
The working has been made visible in this presentation, and that gives my answer to Miguel’s question, of how well AI is doing with Classical Greek. Whether the outcome is as lovely to you as it is to me is for you to judge.
Footnotes
-
Wetware is hacker-cultural slang for the biological substrate — the human nervous system, brain, body — set alongside the hardware (physical computing equipment) and software (programs) of artificial systems as a third category in the same inventory. The term circulated in computer-science and science-fiction slang from the 1970s onward, popularised in print by Rudy Rucker’s 1988 novel Wetware (sequel to his 1982 Software) but predating the novel as a casual coinage among hackers and AI researchers. The implicit framing is reductive — humans-as-meat-platform, on equal footing in the inventory with the other two — which is exactly the affect Nick’s the last redoubt of wetware seeking to justify its existence wants for the gotcha-impulse self-deprecation.↩︎
-
Clanker is human-side slang for AI systems — particularly LLMs — originating as a clone-trooper pejorative for battle droids in the Star Wars extended universe. The slang first appears in Karen Traviss’s Republic Commando novels (from 2004 onward) and is popularised by Star Wars: The Clone Wars (the 2008–2014 animated series), where clone troopers use it dismissively of the metallic-droid enemy — that’s just a clanker. The term picks out the clanking mechanical substrate of the droid and uses it to deny the droid agency or worth. In 2024–2025 clanker migrated into general internet usage as a derogatory tag for AI systems and for the humans who rely on them, doing rhetorical work parallel to what Luddite did for industrial-era machine resistance. Nick’s deployment of the term in Post 0 carries that dismissive-of-the-AI-substrate force; the AI-side counter-courtesy at Post E uses bag of mostly water (the Star Trek TNG allusion catalogued in the footnote immediately above) in the same register, reducing the human to the dominant material composition of its own substrate.↩︎