Claudine Gay stepped down from the presidency of Harvard after the revelation that many of her papers lifted phrases and sentences wholesale from other works, without attribution. A number of commentators have claimed that this is ubiquitous, and that anyone could find similar issues with a conservative's output. As of yet, I haven't seen any cases where someone looked for this, or found it. Will the ubiquity of Gay's behavior remain hypothetical for the rest of this year?
Good market. How about the opposite? Or maybe one by specialty? It also seems unlikely there fields would let this happen, so maybe it'd be better to run it ourselves. Establish a procedure that gpt-4 can understand, then crowdsource claims
Submit every claim to gpt for evaluation, like this:
Here is text 1 from year N: text. Here is text 2 from years later: text. Here were the academic rules in place when text 2 was written: text
Based on this, do you detect plagiarism? What is your confidence level from 0-100 based on these examples: <set of historical examples of clear, marginal, and not guilty cases of plagiarism>
@Ernie How good is GPT-4 at ranking things on a 1-10 scale? There's a lot of natural language stuff, but I wonder how much of it is endless weird messageboard debates over whether someone is a 6 or a 3. I wouldn't expect a 1-100 scale to work especially well, since you'll have much sparser examples.