AI Roundup March 13, 2024
Do you get rewarded for learning AI-generated Torah? AI Roundup called open-sourcing Grok a week ago; Future queen tries her hand at Stable Diffusion
Can we fill in the rest of the Bavli and Yerushalmi this way?
The other day I sat down with my boss for our weekly meeting and being that we both have an interest in AI and Torah he often shares with me things that are put out on one of his messaging groups about AI and Judaism (I would join but the messaging firehose is just too strong these days). Anyway, he told me that he had just received a 16th perek of Mesilat Yesharim which was generated after feeding a Hebrew LLM the entire text of the Mesilat Yesharim and asking it to write one more chapter. The name of the chapter is the “Midat HaAchdut” or the “Virtue of Unity”.
The real question I have is if learning this “perek” counts towards Torah learning time. On the one hand, it is probably no different than some of the divrei Torah you see published by real people that make no sense, or are merely drawing the target around the arrow. Presumably, these types of Torah would fall under the category of Torah learning despite their uselessness because the thoughts were based off of other Torah that the person knew. On the other hand, the LLM has no real knowledge of Torah, rather it just predicts what would be said. It has no holiness to it because it is disconnected from the chain of the mesorah. It isn’t commenting on the Mesilat Yesharim, it is trying to fake its way into the cannon of our heritage.
But if we do decide it is real Torah, where else can we apply this? We are missing almost all of Seder Zeraim in the Bavli. Can we somehow recreate that by seeding an LLM with the Mishnayot? Would it come up with any conclusions/decisions we haven’t already thought of?
Grok goes Open Source: You heard it here first
It was really Musk’s only option after calling the kettle black last week in his lawsuit against OpenAI. I still think that paywalling it completely will be the biggest deterrent to getting people onto his platform, unless through the open source it becomes that much better than ChatGPT4 and Gemini. It would need a lot of hype, as well as for people to overcome the Musk-hate for all of that to happen. As with everything, time will tell.
Kate Middleton Learns Stable Diffusion the Hard Way
Ok yes, this is totally not what happened, and it probably wasn’t even the Princess herself who fumbled the Photoshopping, but it is a good lesson for us all.
Count all of the fingers in your pictures (Stable Diffusion prefers 6, Gemini prefers Black George Washingtons)
Make sure all of those fingers are connected to body parts (preferably the correct ones)
Don’t let the intern publish directly to the media. See rules 1 and 2.
We need a whole article on how the different image generators made the images for this post.