Short answer: we don’t know and it seems really challenging that we will, at least in the near term.
LLMs are black boxes and we don’t know how they work. We can’t follow and explain the logic like we have been able to in all software since the dawn of computers. Machine learning started us down this trail and LLMs have made it common. You can’t even get the same answer if you send in the same prompt twice.
I find this a bit concerning. I am not a AI doomer as I don’t think AIs will cause human extinction. I do think that AIs will significant change human culture across the planet. It is the agricultural, industrial, and computer revolution all wrapped up and amplified.
But not understanding how it works is a problem that we need to solve. This article goes into great detail about how they are trying to do this. Much of it goes over my head, but I can understand that they have tried many things and none of them have worked.
I write science-fiction and in the backstory of one of my novels is that part of the galaxy (the part doing well) banned AI until they could make it a white box. None of the story is dependent on this fact, but I put it in there anyway. Maybe it will impact a future story.
https://open.substack.com/pub/astralcodexten/p/god-help-us-lets-try-to-learn-about
Leave a Reply