Just published: My article in the Seforim Blog on digital layouts of the Talmud - and on the possibility of an AI chavruta
The traditional tzurat hadaf; The digital daf; Sefaria and Al-Hatorah; an artificial intelligence-powered chavruta
A new piece of mine on digital layouts of the Talmud appeared in the Seforim Blog!: “From Print to Pixel: Digital Editions of the Talmud Bavli”.
I discuss the traditional tzurat hadaf, the digital daf, and review a number of layouts, such as that of Sefaria, Al-Hatorah, and many more.
After reviewing a number of contemporary digital layouts of the Talmud, I write:
[A] new possibility has emerged that was not even envisioned a year ago: an artificial intelligence-powered chavruta. I recently experimented with ChatGPT by providing it with the first paragraph of Kidushin 2b.[1]
As usual, ChatGPT demonstrated impressive coherence and confidence. However, it provided laughably incorrect interpretations. In fact, in some cases, it provided interpretations that were the exact opposite of the correct one.
I conducted a comparable test by inputting a paragraph of Ramban's commentary on that sugya, requesting an explanation from ChatGPT. To my surprise, ChatGPT provided a reasonably good rephrasing in English. Although some of it was correct and some of it was incorrect, even the incorrect part was not entirely off. However, ChatGPT still requires further yeshiva study before being able to serve as a teacher.[2]
Given the rapid advancements in artificial intelligence and large language models, it appears highly likely that AI will continue to improve its ability to interpret the Talmud, along with all other sources, including digitized manuscripts. It may not be long before we turn to AI to obtain the ultimate p’shat in the Talmud.[3] Furthermore, there is a possibility that AI could eventually bring comparative sources to bear, as previously mentioned.
To update the speculative future envisioned by Fischer and Secunda, my ideal Talmud study experience would be an app with a GPT plug-in acting as a virtual chavruta/Rebbe, to interpret text and answer questions. In fact, something like this is already being tested by Khan Academy.[4]
With current tools, it will even be possible to create a virtual shiur.[5] In the future, it could even be live and interactive, happening in real time.
See much more there, at the link. Self-recommending!
[1] On 15-Apr-23, at https://chat.openai.com/
[2] I should point out that at the time, I did not have access to GPT-4, which is the latest version of GPT. So I could not test whether GPT-4 is more capable at interpreting Talmud. For a discussion of some sources of the Yeshivish dialect of English possibly used as datasets for training ChatGPT-4, see my “From the Shtetl to the Chatbot: Some contemporary sources of Yeshivish content, in light of ChatGPT-4”.
[3] For some preliminary algorithmic research on the Talmud, see, for example, Satlow M., Sperling M. (2017). “Naming Rabbis: A Digital List”; Satlow M., Sperling M. (2020), “The Rabbinic Citation Network”; Satlow M., Sperling M. (2020). “The Rabbinic citation network”, AJS Review ; Zhitomirsky-Geffet M., Prebor G. (2019), “SageBook: toward a cross-generational social network for the Jewish sages’ prosopography”, Digital Scholarship in the Humanities, 34(3), pp. 676–695 ; “A graph database of scholastic relationships in the Babylonian Talmud”, Digital Scholarship in the Humanities, Volume 36, Issue Supplement_2, October 2021, Pages ii277–ii289, https://doi.org/10.1093/llc/fqab015, Published: 22 February 2021. See also Shira Shmidman, “Self-Evident Questions and Their Role in Talmudic Dialectic”, AJS Review (2023), p. 128: “Recently, I analyzed talmudic questions (that open with the phrase baʿei or beʿah minei) and discovered that by charting the questions by the generation in which they were asked, one can identify chronological trends of question asking in the Babylonian Talmud.”
[4] See Sal Khan, “Harnessing GPT-4 so that all students benefit. A nonprofit approach for equal access”, Khan Academy Blog (March 14, 2023). See especially this description: “Khanmigo [=the AI tool] engages students in back-and-forth conversation peppered with questions. It’s like a virtual Socrates, guiding students through their educational journey. Like any great tutor, Khanmigo encourages productive struggle in a supportive and engaging way.” For fun, I asked ChatGPT for a clever name for an AI chavruta/Rebbe. It gave me ten options. One of them is a great one (ChavrutAI), while three of them were hilarious ( AIvrumi; AIsh Torah; RoboRabbi). RoboRabbi could provide many of the intellectual aspects of being a rabbi: psak, eitza, questions in learning, and divrei torah.
[5] With tools like ChatGPT (content), ElevenAI (voice), D-ID (video), one can create any style of shiur. See my “Heimish High-Tech: Video in Yeshivish dialect using Generative artificial intelligence”. For training sources for the Yeshivish dialect, see my article: “From the Shtetl to the Chatbot: Some contemporary sources of Yeshivish content, in light of ChatGPT-4”.