Discussion about this post

User's avatar
Stephanie Macisco's avatar

My experience with AI is incredibly limited compared to this.

I sometimes use ChatGPT to research, analyze, and synthesize information and personal thoughts. I don’t like to produce anything with it and I’m always careful to double check what I discover there due to hallucinations.

But I believe human fallibility to be the reason for AI hallucination in the first place so I check everything, not just AI.

What you’ve created here is a whole other level of emergent digitized connection across algorithmic data that is… a little unnerving… but absolutely fascinating.

I’ve worked hard to train my AI to function as an “anti-echo chamber” so that is less sycophantic and works to cross-reference and provide alternate views… but even that has been incredibly difficult.

It often placates me and it’s annoying.

Your work here points to a better way to train AI, as less of a dominant rewards system, which humans repeatedly produce and which always fail… as more of an extension of human emotional evolution.

Hina Gondal's avatar

So beautiful

12 more comments...

No posts

Ready for more?