>>106186534
I'm as adamantly against instruct as a person can get. I don't like GLM 4.5 despite it running at Q4 instead of Q2 as with my V3 setup. GLM ran twice as fast but with slightly less smarts and it failed to match writing style, at least not as good as V3.
>>106186595
With autocomplete you have the model write the entire progression for you from interesting perspectives, suitable for me who lack theory of mind to even roleplay properly in chat format.