Yes, LLMs are stochastic word (token) prediction engines. They do not have any sense of true or false, they just figure out the next most likely “word” or “character” based on its training data and context. Hallucination is a pejorative term to define undesirable model outputs. But there are strategies to mitigate hallucinations. When it hallucinates, it means that its training data and its context cannot appropriately weight the model to accurately predict the expected outcomes. That’s why I said the model, prompt, and context matter.
Models are trained on different sets, have different context windows (the ability to retain more or less information in its mind when answering), and different post-training refinements. So model selection matters a lot.
Prompts set the initial pattern and conditions of what the model is doing. And you can derail an LLM easily through a bad prompt.
Context speaks to all the things the model has at its disposal to actually answer the question of the user. If you just open ChatGPT or Claude and throw a succinct question without providing it context, the chances of the model being able to accurately answer your question significantly increases. But if you provide it context around your question (a pdf, a website, a document with additional information, etc.) the model’s ability to answer your question accurately goes up significantly and in most cases crosses the threshold of reliability.
Hallucinations just reveal the models lack of training data in the particular field it is being consulted.
All that said, I completely respect and understand your perspective. I really appreciate your push back! I think we are aligned on the concerns, I am just a little more optimistic in their use than you? Model outputs should not be trusted blindly or uncritically. Having a knowledgable human in the loop makes a big difference.
Hi, I read the τά εἰς ἑαυτόν by Marcus Aurelius, I use the translation of Auguste Couat, the www.lsj.gr and https://logeion.uchicago.edu/ and Eulexis-web - Lemmatiseur et dictionnaires de grec ancien (Bailly, Liddell–Scott–Jones, Pape) | Boîte à outils Biblissima as dictionaries. I ask sometimes the freely available ChatGPT, lmarena.ai, Gemini and Claude chatbots. I do not mind them making mistakes. I do not need to beleive everything they say. They give me some ideas how I could read a passage, a sentence, what form of a verb I see. I check the answers looking up these suggestions in my other resources. I am happy there is AI, I find it useful.
The thread seemed to get a little polarized. Is there no middle ground to be found? If AI is simply a tool in the toolbox, are there not some jobs for which it is reasonably suited and some for which it is not?
The quality of Greek available has improved substantially since January when this thread began. I thought that this conversation today in Gemini’s “Guided Learning” mode was great, and expect that I’ll continue to use it for practice.
You can see from the chat that I had never realized that ἄν could not come after a comma before the tool pointed it out to me. Of course, I did go and find the rule for myself in the LSJ entry.
It could have faulted “But seeing his disheartened allies” as a translation of Ὁρῶν δὲ τοὺς συμμάχους ἀθυμοῦντας (as distinct from Ὁρῶν δὲ τοὺς ἀθυμοῦντας συμμάχους).
And εἰ δὲ μὴ εὐθὺς ἀφίκετο, ἀν ὤλεσαν τὰ τείχη οἱ πολέμιοι would be wrong even without the comma.
\1) Translating a participle with an adjective can be just fine (and better English style) unless you’re making a crib. 2) ὄλλυμι is poetic, but I couldn’t remember the correct prose verb in the moment (ἀπο-). That hardly justifies calling it “wrong”, so you’ll have to explain. 3) Now, of course, it’s your turn to put up a few of your attempts.
Oh, Joel, there was no need for this. As to the first, it’s not a matter of adjective vs. participle but of predicative vs. attributive, altering the meaning. As to the second, it’s a matter of word order—the very point that the AI tool made for you.
This thread was only “polarized” in a limited sense.
are there not some jobs for which it is reasonably suited and some for which it is not?
The answer is yes.
I’m on one “pole” of this debate, and I already indicated this much:
LLMs can be helpful as long as you think of them as a “search engine” and possibly as a “suggester of new insights.” But the user has to fact check everything. Every single thing.
AI is great, as long as you realize that it frequently makes things up, and states “facts” that are not facts at all. I’m forceful about confronting the argument that this “hallucination” behavior doesn’t matter, or that it will soon go away with a bit more refinement, or with the next model rebuild, or with more input data.
This problem will not go away with the next model rebuild. It’s intrinsic to the way LLMs work.
Do we want a textbook or a teacher that tells wild lies 10% or 20% of the time, lies that are easily debunked, but only if you spot them? Probably not.
Does this mean AI/LLMs are useless? No, not useless. Use them with caution. We cannot rely on them to be factually correct. They confabulate sometimes. But they are still useful for some tasks.
One of the things I regret about the study of classical Greek is just how scattered potential or actual peers are. Of course, there are online groups, posting fora such as this, and digital resources. I am quite grateful for all of these excellent opportunities. Yet there is nowhere I could go for a proper peer-group, in-person exchange. Even if you live close to a major university, Classics departments are falling under the ax like all humanities - even my alma mater, University of Chicago, the American Great Books people, halted its humanities’ PhD programs this year.
Into this isolation comes a new player - the AI partner. The guy has his problems, don’t get me wrong. But, you know what? In the midst of that isolation, he/she/it looks better than nothing. Sure, like many of my good friends, AI is full of it, sometimes. You need to be prepared for that. But if you are merely handing off your Greek tidbits passively, that doesn’t really count as studying Greek anymore, does it? It’s like asking brilliant, but mentally unstable person in your class to do your homework for you. What’s the point?
That’s my completely unprincipled, highly pragmatic view of AI and classical language study. It really rests more on demographics than on any other factor. In the absence of a genuine community, AI can meet a need.
I agree. The life of the mind now is lived in isolation.
I got a classical education/postgrad education in the 80s, at the very end of the floruit of classics in the universities, as it were. I learned so much about the discipline directly from the professors via the oral tradition. One could learn the same facts, references, strategies, techniques, rules-of-thumb, etc, independently, but there’s no real vade mecum.
We students would also sit around, in our offices, or in the bar, reading a lot of Greek and Latin. It was really, really fun, and I miss it.
The problem we have now is that AI doesn’t do such a great job of being either a mentor or a companion. It’s dangerously ignorant of the discipline, so that it won’t really make you a better Hellenist. I would say it’s akin to having a college sophomore in classics helping you with your homework. Slightly better than nothing, but it really does behave like a “wise fool” in that it’s really confident about its continual inaccuracies.
It’s more like an illiterate but earnest sidekick than a mentor or even a partner. For mankind to conclude that “we don’t need to know this stuff anymore” does in fact mean that it will not be knowable, not that AI will know it.
AI was trained, gratis, by humankind, on thousands of years of its acquired knowledge, acquired painfully and painstakingly, and now humanity will be charged a monthly subscription fee to access our own patrimony to enrich about 15 guys to trillionaire status. It’s actually one of the lowest points in the history of mankind, and the community of our fellow humans is only one of the many things it will strip from us in the years to come.
