A creative team has released an interactive shiur on Basi Legani 5724 – developed fully by AI. Are they torah-ically correct? or they’re gibberish? Listen and decide.
By Rayi Stern
‘Evan’ Mikir Tizak
Every year and every day represents a new potential and new capabilities which were never seen before. Today, in honor of Yud Shevat, we’re presented with what can only be referred to as a ‘treasure’ from above, or, as the maamar puts it, an אוצר מלמעלה.
A dedicated team has released an interactive shiur on Basi Legani – developed fully by AI. Much like the printing press, its obvious what purpose Artificial Intelligence is here for. Applying AI to Torah is a no-brainer.
The shiur, available on Spotify, Apple Podcasts, and more, covers the entire maamar from 5724. And it doesn’t just translate – it delves into the important ideas and explains them thoroughly. The team did step in to edit the content, since, of course, one has to be careful when it comes to Torah, and especially with the Rebbe’s Chassidus. There weren’t any major changes to be made, however; moreso nuances had to be addressed, for example, the AI mixed in the term ‘Essence’ when discussing Atzilus. “We also had to discard a ton of content. Simply because its not a 12 part series, its one single shiur,” says Rayi Stern, who was involved in an advisory role.
And its also interactive. Evan, the podcaster, will WhatsApp you back.
One big question, the same question being considered across the whole world, is what is the long term application of this technology? Its obviously not just to write our emails and give us recipes — and not even to just give a shiur.
Evan basically becomes an AI Ask the Rabbi at that point, and he even responds to voicenotes! He has broader content in Chassidus at his disposal, as well the ‘enriched’ content which went into his shiur.
So what does it mean that an AI created a shiur? Of course, asking ChatGPT to do something like that won’t end well. The team tapped expert knowledge in this up and coming field, and designed strategies and automated processes which direct the AI into extracting and developing the key content from within the maamar, and also from other sources within Chassidus. So what we’re seeing in the end result is only about 10% of the total content generated. “There was a lot of back and forth amongst our bots. And its fascinating to look at the questions and explanations which came about.”
Is the AI making it all up? The team says that the output is highly accurate, and routed in the sources provided. “When it comes to Basi Legani, there’s a lot of detail on each point, since there’s so many maamarim to work with.” And they also used industry-standard practices to restrict the AI to the provided sources and to avoid hallucination. “We learnt a lot throughout this process (not just in the maamar, but also in AI). But then again we also came in with experience and expertise.”
What’s the catch? The team admits that there are some shortcomings, and they also ran into many challenges throughout the project. “We were building a system, and we’re proving a concept. Nothing that we did was for this maamar alone.”
As this was a test run for them too, they really pushed the limits of what AI can do right now. “There are certain embedded ideas in the AI, which are hard to sway. Concepts like ‘etzem’, ‘giluyim’, ‘etzem hanefesh’ – certain points we found were tough for it to wrap its AI head around. But then again, humans have trouble there too.”
They plan to expand their platform to include a wider range of AI solutions and technologies, which opens the door to more tailored approaches. “The space is advancing so rapidly; its not about putting out a state of the art product – its about a solid strategy which is the complemented by each new advancement.”
The goal of this project was to open up a discussion about what’s out there, and about what can be done with it. Many would say that today we’re at a time where ‘לא ילמדו עוד איש את רעהו’; where the pursuits of the world are shifting from profit seeking more toward intellectual exploration and social development, etc. The general response, they say, has been wheels turning in people’s heads; how a theoretical system that’s tailored to Torah principles and thought processes is readily available to be produced, using existing technologies. “And that’s what we aimed to point out with Evan.”
AUDIO:
Great article–not a correction per se, but it’s worth noting that AI is basically a marketing term because computer scientists do not have a strong definition of intelligence or consciousness. It’s generally more effective to think of it as automation by means of Large Language Models.