139 - Coherent Long Story Generation, with Kevin Yang
NLP Highlights
00:00
How to Re-Rank Your Outputs With GPT-3
In the rewriting path you have these models trained to evaluate coherence and relevance. You use them to re-rank your outputs can you describe those models in detail please? Yeah so there's two separate re-rankers here there's a relevance re-ranker and a quick-grants re- Ranker. The relevance re-Ranker is designed to make sure that the story continuation actually stays faithful to the plan. And then at inference time we're actually generating our story this re- ranker takes the passage that we just wrote and then compares it to what the plan says we're supposed to do next.
Transcript
Play full episode
