Thank you for not only challenging the status quo around how LLMs function, but also offering a practical alternative beyond RLHF and constitutional AI. I commend your collective for that, however, concerns arise for me with this democratic fine tuning process that I'm happy to engage more about.
The appropriation of wisdom as something that trickles up from democracy and massive public processes is an interesting, but potentially dangerous claim. In a sense, the operator intent is just moving from powerful AI leaders to a larger scale collective intelligence, but are we sure this leads to wisdom? Have massive public processes historically proven to result in wisdom? Of course we don't want AI leaders in charge of fine-tuning these models, but I'd suggest that our collective is also an adolescent culture that is not trustworthy. And there is a unique dynamic wisdom in the collective, so how do we truly bring this out?
Wisdom traditions tell us that wisdom is not created and made up, nor is it something we carefully think up or feel into. Don't misunderstand, I think reflection between large groups that include diverse and marginalized perspectives is important, but this isn't wisdom. Wisdom is the ability to get out of the way and allow something beyond the conditioned human ego, intellect, and value system to move through us, and ideally manifest in the structures of this world for the benefit of beings. AI offers an unprecedented opportunity in this regard.
I'd like to see a process like this, but that is refined towards revering the wise elders, not just the mass collective intelligence. Hierarchy and more siloed power might not necessarily be bad, if those with power have true wisdom and love. So the question that I'm wondering is: how do we offer power to the wise and loving, facilitate collaboration between those people, and find ways to reflect this wisdom and love through mediums of AI and other emerging technologies? What would a fine-tuning process look like that is run by the wise elders? And how do we create models that reflect true wisdom in a way that takes into account pre-existing values, views, and preferences, so that it meets people where they’re at?
Exciting work! Can’t wait to try the prototype 👏🏼
Thank you for not only challenging the status quo around how LLMs function, but also offering a practical alternative beyond RLHF and constitutional AI. I commend your collective for that, however, concerns arise for me with this democratic fine tuning process that I'm happy to engage more about.
The appropriation of wisdom as something that trickles up from democracy and massive public processes is an interesting, but potentially dangerous claim. In a sense, the operator intent is just moving from powerful AI leaders to a larger scale collective intelligence, but are we sure this leads to wisdom? Have massive public processes historically proven to result in wisdom? Of course we don't want AI leaders in charge of fine-tuning these models, but I'd suggest that our collective is also an adolescent culture that is not trustworthy. And there is a unique dynamic wisdom in the collective, so how do we truly bring this out?
Wisdom traditions tell us that wisdom is not created and made up, nor is it something we carefully think up or feel into. Don't misunderstand, I think reflection between large groups that include diverse and marginalized perspectives is important, but this isn't wisdom. Wisdom is the ability to get out of the way and allow something beyond the conditioned human ego, intellect, and value system to move through us, and ideally manifest in the structures of this world for the benefit of beings. AI offers an unprecedented opportunity in this regard.
I'd like to see a process like this, but that is refined towards revering the wise elders, not just the mass collective intelligence. Hierarchy and more siloed power might not necessarily be bad, if those with power have true wisdom and love. So the question that I'm wondering is: how do we offer power to the wise and loving, facilitate collaboration between those people, and find ways to reflect this wisdom and love through mediums of AI and other emerging technologies? What would a fine-tuning process look like that is run by the wise elders? And how do we create models that reflect true wisdom in a way that takes into account pre-existing values, views, and preferences, so that it meets people where they’re at?