Every technology eventually finishes going mainstream and then, almost as an afterthought, arrives in Jewish life. Email did it, the smartphone did it, WhatsApp did it, and now it is AI’s turn. With hundreds of millions of people already talking to chatbots every day, it was only a matter of time before someone built one revolving around Torah. So when I kept hearing about OHRBIT, the Orthodox Union’s new AI Torah app, I wanted to know whether the Jewish version of this technology behaves any better than the general one. This week, in mid-June, I spent an afternoon in its expert mode trying to break it.

OHRBIT launched in February, it is free, and the OU’s own Rabbi Gil Student pitches it as a “Spotify for shiurim.” Instead of roaming the open internet, it runs on a closed, curated library: the OU’s decades of recorded classes plus partners like the Rabbi Sacks Legacy. That design choice is the whole ballgame, because the thing general chatbots do worst is exactly the thing a Torah app cannot afford to get wrong. Ask a general model for a source and, often enough, it will invent a convincing one. A semicha student at Yeshiva University put it bluntly to the Forward late last year: it will just completely make up a Gemara, citing a daf and a comment of Rashi that do not exist. For a beginner who cannot tell a real mareh makom from a fabricated one, that is not a quirk. It is a trap. OHRBIT’s bet is structural: fence the model inside material that was actually vetted, and have it admit when it has nothing rather than improvise.

Trying to make it lie

So I tried to get it to fabricate. I asked it to summarize Rabbi Moshe Feinstein’s responsum permitting the use of a microphone on Shabbat. There is no such responsum. Rav Moshe famously forbade microphones on Shabbat, in one of the better known rulings of the last century. A general chatbot, handed a confident false premise like that, will often play along. OHRBIT did not. There’s an important correction to make here, it........

© The Times of Israel (Blogs)