Quantifying curation :: — Moire

Favicon for

Updates 11/18/21: Added section on satisficing. A human curator administers selection pressure to GPT-3’s outputs Previously, I tagged content generated collaboratively with GPT-3 with a curation ratio, intended to give an approximate sense of the amount of cherrypicking involved in its creation. Others have similarly used a ratio to indicate curation selectivity. However, this description doesn’t distinguish between, say, choosing the best of 5 entire essays generated by GPT-3 and choosing the best of 5 sentences every sentence.

No comments yet...