I think this has been going on well before AI or LLMs: "A fundamental danger, as we’ve discussed, is that AI can allow students to bypass learning. Equally concerning is that students may internalize a transactional model in which assignments are outputs, teachers are evaluators, peers are optional, and knowledge (or an MIT degree) is an optimizable commodity to be acquired or produced as efficiently as possible."

Ok, but did AI and LLMs make the cheating easier, faster, harder to detect? Whatabout the past doesn't change the problems now.

"My leg already had a cut on it. No need to worry about the new stab wound in my chest!"

Absolutely; it's a real problem. But institutions like MIT have also been participating in, and benefiting from, this transactional relationship from their side in offering degrees that they know carry so much value.

MIT makes all its money from research (and investing its endowment more successfully than most schools), not tuition, and the success in research comes from having such a high-achieving community of faculty and graduate students. They're pretty incentivized to keep the bar for academic achievement genuinely high.

To clarify, I am pro-university. I'm just saying that institutions do not always behave in ways that are purely for educational benefit, such as with legacy admissions or grade inflation. I don't think the blame falls solely on students for viewing university degrees as a necessary step for something else. I would sure bet admissions committees will use AI to screen applicants, if they aren't doing so already.

The French tried this: https://en.wikipedia.org/wiki/French_Republican_calendar

I wonder what they would have thought of swatch internet time.

I think there's more to it than that. It's a combination of extracting the text from the PDFs, resolving persistently tricky issues (especially numbers), and flagging issues for human review at scale. You can view lessons from the author on the GitHub repo: https://github.com/sweisman/translation-pipeline

Great, then rather post that! Although that repo looks like 100% AI artifact, not human intervention. (What does "flagging issues for human review at scale" mean in this context?)

I was referring to this bullet under Lessons Learned: "Manuscript translation is feasible but rough. A 14th-century Sephardic manuscript hand is readable at ~70–80% accuracy. The output is a useful rough draft, not a polished translation. Plan for a longer discrepancy report and more editorial review." Some of the samples I looked at had question marks or original Hebrew text throughout, which I took to mean marked uncertainties.

Manuscripts are doable but tougher than printed works. Hebrew likewise is tougher than Latin.

This is not what I use to translate. It was the first version to create usable output, and intentionally implemented in a code-free manner (today I have an extensive body of code in addition to the Claude Code harness instructions). Read my origin story, linked to in another comment, for more background.

Reiterating what I wrote above, sharing these would have been more interesting imo than sharing the pipeline output.