Comment by Xcelerate

Are there any articles describing how GPT-4 without guardrails is different than the public version? I would be super interested in what the model’s actual “most probable next token” looks like.