Another good article! Two points, the first one you probably can't answer for conflict of answer reasons:
1. on "democratic force of a LLM" this value embedding is why a bunch of my colleagues and I were worried about constitutional AI. It seemed like a better way to encode the values (rather than just scraping text), but doing so at all as a private company didn't seem like something the information ecosystem was ready for. I'd like to keep learning about this (and to see more experiments on RLAIF, they seem less numerous than I would expect).
2. I wonder how many people out there actually understand all of the implications of the word "Agent". Many, many people are going to start encountering it in everyday life and I find that wild. We'll see where it goes!
Another good article! Two points, the first one you probably can't answer for conflict of answer reasons:
1. on "democratic force of a LLM" this value embedding is why a bunch of my colleagues and I were worried about constitutional AI. It seemed like a better way to encode the values (rather than just scraping text), but doing so at all as a private company didn't seem like something the information ecosystem was ready for. I'd like to keep learning about this (and to see more experiments on RLAIF, they seem less numerous than I would expect).
2. I wonder how many people out there actually understand all of the implications of the word "Agent". Many, many people are going to start encountering it in everyday life and I find that wild. We'll see where it goes!