139 - Coherent Long Story Generation, with Kevin Yang
NLP Highlights
00:00
The Power of Re-Rankers in Story Generating
The re-rankers are actually the only part of this whole pipeline that uses the previous story data. So everything else where we were generating a story was operating like all zero shot which is kind of nice stuff that actually trains something and in principle it can generalize as long as you like modify your prompts appropriately. The editing module seems to be more about ensuring some sort of a global consistency rather than local or global consistency.
Transcript
Play full episode
