"The AI is a sycophant" is a much more general problem, as you've noted, but this was a great outlining of its strengths and challenges. I'd add one more (general) problem, which is that all the current models get very easily confused after a few minutes/few thousand words of conversation and cannot maintain the logical flow well. Try reading a Tosafos (or lengthier give-and-take in the Gemara) and see how quickly it loses the thread. Related: with all the talk on AI as a tool for Torah learning, I still have not seen any rabbinic writing on using AI for another, timely intellectual mitzvah: sippur yetziyat Mitzrayim. Perhaps not expected from Orthodox sources, since they'd disapprove of using any technology on the holiday, but have you seen any rabbinic discussion of this at all?
Great post, and reflects very closely what has been my experience using AI for Torah-related research. I also strongly agree with your last paragraph on the importance of human interaction for a chavrusa. It will be very interesting to see how the technology improves (or doesn't?) in this small corner of AI over the next 5-10 years.
One area of Jewish learning that I believe ChatGPT will have a major impact on very rapidly is translations. I have found that copying and pasting text, regardless of whether it is in Hebrew or Aramaic (or a mix, or even Yiddish), into ChatGPT and asking it to translate it, produces very impressive translations. It is even adept at deciphering Roshei Tavos. You can also use Google Lens to copy and paste text from an actual book into ChatGPT with the same result (although it requires a few more steps than simply copying and pasting on a desktop). It is only a matter of time (2-4 years?) before Sefaria or AlHatorah integrate ChatGPT-driven translations into their interface. I wonder what kind of effect this will have on the drive to learn how to read texts in their original language. Thoughts?
"The AI is a sycophant" is a much more general problem, as you've noted, but this was a great outlining of its strengths and challenges. I'd add one more (general) problem, which is that all the current models get very easily confused after a few minutes/few thousand words of conversation and cannot maintain the logical flow well. Try reading a Tosafos (or lengthier give-and-take in the Gemara) and see how quickly it loses the thread. Related: with all the talk on AI as a tool for Torah learning, I still have not seen any rabbinic writing on using AI for another, timely intellectual mitzvah: sippur yetziyat Mitzrayim. Perhaps not expected from Orthodox sources, since they'd disapprove of using any technology on the holiday, but have you seen any rabbinic discussion of this at all?
Great post, and reflects very closely what has been my experience using AI for Torah-related research. I also strongly agree with your last paragraph on the importance of human interaction for a chavrusa. It will be very interesting to see how the technology improves (or doesn't?) in this small corner of AI over the next 5-10 years.
One area of Jewish learning that I believe ChatGPT will have a major impact on very rapidly is translations. I have found that copying and pasting text, regardless of whether it is in Hebrew or Aramaic (or a mix, or even Yiddish), into ChatGPT and asking it to translate it, produces very impressive translations. It is even adept at deciphering Roshei Tavos. You can also use Google Lens to copy and paste text from an actual book into ChatGPT with the same result (although it requires a few more steps than simply copying and pasting on a desktop). It is only a matter of time (2-4 years?) before Sefaria or AlHatorah integrate ChatGPT-driven translations into their interface. I wonder what kind of effect this will have on the drive to learn how to read texts in their original language. Thoughts?