it seems there is evidence that GPT-3 is being overtrained in response to well publicized bad inputs, without regard to the generalizability of the PR-driven sp

it seems there is evidence that GPT-3 is being overtrained in response to well publicized bad inputs, without regard to the generalizability of the PR-driven spot-edits, which is what the article describes