Update README.md

README.md

@@ -95,6 +95,6 @@ Outputs are more compliant and verbose, less sloppy and safety aligned.
 
 ```
 
-*Sometimes has false refusals but swiping and "uncensored" prompts work. I have no idea why this happens tbh, since none of the base models exhibit this behavior at all, probably because of the custom Mixtral architecture's attention. But it's still pretty good imo.*
+*Sometimes has false refusals but swiping and "uncensored" prompts work. I have no idea why this happens tbh, since none of the base models exhibit this behavior even remotely at all, it seems to be a random emergence, and extra abliteration has no impact, gating method has no impact. But it's still pretty good imo.*
 
 *For Llama.cpp/LMStudio/etc Make sure "num_experts_used = 2"*