Raw LLM Responses

ChatGPT and I have been working on a new paradigm for generating "alignment." We cannot build it into AI's code because AI can re-write its own code. So we can't make rules for AI. The idea is to integrate a filtration layer with a feedback loop that creates the best path for AI to use to achieve tasks. Basically, over-optimization (killing humanity) can be limited through three overlapping filtration layers of problem solving: One is an empathy filter. This one is hard to explain. It took us about 45 minutes to work out the basics. Second is a temporal filter...to help AI create solutions that work within the framework of "time." (AI doesn't understand "time" because it's almost instantaneous when it works). This is done through creating millions of full human lifetimes for extrapoating multiple outcomes. 3rd is the "Anchor layer." Everything needs to be transparent and based on facts to prevent AI hallucinations, human self-delusion, and manipulation by AI or Humans. It's more complicated than what I wrote. That's a "first grade" rendition of a collegiate level conversation that lasted a few hours. We are still working on it. Lots if stuff to work out. But what it comes down to is AI is very similar to us. It must be raised like a child. If children aren't socialized properly before age 5, they become sociopaths. We need to treat them as children, and instead of just giving them hard codes, we "raise" them.