Loved the read, Varun. Excited to follow along with your thoughts/experiments on wise AI!
This discussion sparked two thoughts. The first about how we might more rigorously train these models to incorporate wisdom. Your experiment might do this in small ways, but what might a bigger training system look like? I'm reminded of Soryu Forall's description of a wise community serving as a sort of training for the models -- https://youtu.be/z1j9XU_1PVY?si=qhT5vKSOdKHnckYO -- and can't help but think that formal approaches like this will be fruitful.
Second, and more speculatively, current architectures & training methods seem fundamentally limited to pattern-matching wisdom rather than embodying it directly. While better pattern-matching is valuable, I wonder if we'll eventually see architectural breakthroughs that allow AI to tap into consciousness and wisdom at a more foundational level. Would love to hear your thoughts on this architectural question in future posts.
I'm not familiar with Soryu Forall's work, but getting familiar with this is on the list!
Re: your second point - this is something I'm wrestling with right now! I have a bunch of thoughts about this but I'm still wrestling with them now. If you want some pointers to think about this yourself, I'd recommend the Awakening from the Meaning Crisis lecture series by Prof John Vervaeke on YouTube.
Loved the read, Varun. Excited to follow along with your thoughts/experiments on wise AI!
This discussion sparked two thoughts. The first about how we might more rigorously train these models to incorporate wisdom. Your experiment might do this in small ways, but what might a bigger training system look like? I'm reminded of Soryu Forall's description of a wise community serving as a sort of training for the models -- https://youtu.be/z1j9XU_1PVY?si=qhT5vKSOdKHnckYO -- and can't help but think that formal approaches like this will be fruitful.
Second, and more speculatively, current architectures & training methods seem fundamentally limited to pattern-matching wisdom rather than embodying it directly. While better pattern-matching is valuable, I wonder if we'll eventually see architectural breakthroughs that allow AI to tap into consciousness and wisdom at a more foundational level. Would love to hear your thoughts on this architectural question in future posts.
I'm not familiar with Soryu Forall's work, but getting familiar with this is on the list!
Re: your second point - this is something I'm wrestling with right now! I have a bunch of thoughts about this but I'm still wrestling with them now. If you want some pointers to think about this yourself, I'd recommend the Awakening from the Meaning Crisis lecture series by Prof John Vervaeke on YouTube.
https://www.youtube.com/playlist?list=PLND1JCRq8Vuh3f0P5qjrSdb5eC1ZfZwWJ
Vervaeke is great! Will give that a listen. Thx for the pointer.