ChatGPT and most AI writers can only generate about 500 ~ 3000 words. Infklow.io can draft 25,000+ word E-Book in a few minutes. (Built for Non fiction - Speed up the first draft)
I’m not watching a video that should be a text description of a technique, but the obvious solution for the 4096 output token limit on the 128k token context GPT-4-Turbo is to run calls to the model in a loop where the output for the previous run is added to the input with instructions to continue from where it left off.
While GPT-4 or other SOTA LLMs can probably be good tools for help writing a book, generating a whole ebook with them (or even really long stretches without human input and tweaking and reprompting) is probably a really bad idea with current SOTA quality, and working around the output size limit on the models that have one is probably the least difficult part of resolving that.
6 comments
[ 2.8 ms ] story [ 18.8 ms ] threadWhile GPT-4 or other SOTA LLMs can probably be good tools for help writing a book, generating a whole ebook with them (or even really long stretches without human input and tweaking and reprompting) is probably a really bad idea with current SOTA quality, and working around the output size limit on the models that have one is probably the least difficult part of resolving that.