Deprecation notice: This method of approximating quantity of curation used only for old content (authored < May 2021). For newer content, I use exact metrics.

When I post things that cowritten by GPT-3 or other bots, I will label the content with two ratios, contribution and selectivity. If there is a significant change in the balance for certain sections of the post, I may restate the scores.

Contribution [AI-authored text : human-authored text]. Approximate proportion of AI generated to human-written text in the final product. “labelled inline” means that I indicate all human/AI contributions inline via formatting or otherwise.

Selectivity: Approximate proportion of AI generated text included in final product to total AI generated text out of which it was selected (size of branch / size of multiverse); approximately equal to branching factor.

These ratios are usually approximate. Some kinds of human contribution such as deleting or moving text or hidden prompts are not captured by literal interpretations of these measures, so when I make significant use of such modifications, I’ll fudge the contribution factor to better reflect the how much control I exerted over the text.

Example

This means that about 90% of the text that appears in the post was written by GPT-3 and 10% by a human, and that the GPT-3 written text in the post represents about 20% of the total GPT-3 generated content out of which it was selected.