This Generative Ink post talks about curating GPT-3, creating a much better output than it normally would give, turning it from quite often terrible to usually pround and good. I’m testing out doing the same with this post, choosing one of many branches every few dozens of words.
For a 4x reduction in speed, I’m getting very nice returns on coherence and brevity. I can actually pretend like I’m not a terrible writer! Selection is a powerful force, but more importantly, continuing a thought in multiple ways forces you to actually make sure you’re saying things in a way that makes sense.
Editing my writing can be slow and tedious, but this excercise gives me a way to naturally write in a compact, higher-quality way, which is why I hope I do this many times in the future, and recommend you try doing this yourself.
This post was written in 5 blocks, and I wrote 4 (= 2^2) branches for each block, for 5*2 = 10 bits of curation, or 14.5 words per bit of curation.
As it happens, I always used the final branch for each block, so it was more effects of revision and consolidation than selection effects that contribute to the end result of this excercise.
This Generative Ink post talks about curating GPT-3, creating a much better output than it normally would give, turning it from quite often terrible to usually pround and good. I’m testing out doing the same with this post, choosing one of many branches every few dozens of words.
For a 4x reduction in speed, I’m getting very nice returns on coherence and brevity. I can actually pretend like I’m not a terrible writer! Selection is a powerful force, but more importantly, continuing a thought in multiple ways forces you to actually make sure you’re saying things in a way that makes sense.
Editing my writing can be slow and tedious, but this excercise gives me a way to naturally write in a compact, higher-quality way, which is why I hope I do this many times in the future, and recommend you try doing this yourself.
It occurs to me that this is basically Babble & Prune adapted to be a writing method. I like Babble & Prune.
This post was written in 5 blocks, and I wrote 4 (= 2^2) branches for each block, for 5*2 = 10 bits of curation, or 14.5 words per bit of curation.
As it happens, I always used the final branch for each block, so it was more effects of revision and consolidation than selection effects that contribute to the end result of this excercise.