I guess what I was trying to illustrate that if you train an LLM with RLHF, the analogy is squeezing the directionless network along a specific axis, but then you get both the friendly face and the evil face, two sides of the same coin squeeze.
I guess what I was trying to illustrate that if you train an LLM with RLHF, the analogy is squeezing the directionless network along a specific axis, but then you get both the friendly face and the evil face, two sides of the same
coinsqueeze.