[dbingham]

I honestly have no idea. I don't think I know nearly enough about GPT3 to hazard a guess.

I could imagine using natural language processing as part of looking at generating an automated Q&A algorithm or attempting to automate literature reviews in some way, but in the review process?

Someone I was talking to the other day was suggesting using sentiment analysis during the review process as a kind of tone grammarly aid to help people write constructive reviews, which is interesting. But I think that's different from GPT3.

Judging from your bio, I would guess you have much strong ideas about that answer to that question than I do. What are your thoughts?

[dr_dshiv]

If you like, you can link me to a paper you know well and I’ll send a review based on gpt3, without reading it. You can tell me if it is sensible.

I rather like the frontiers review process as a gatekeeping process. Papers get much better through their interactive review. But I don’t think peer review should stop with publication. I think there is a need to rate and rank and otherwise gather sentiments from researchers on papers in their field —ideally in a manner that allows for new important work to surface more easily. Whether the goal is to make science human and machine readable, for the further advancement of science. There is going to be a lot of science.

* check this out: https://www.nature.com/articles/d41586-022-02787-5

[dbingham]

> I think there is a need to rate and rank and otherwise gather sentiments from researchers on papers in their field —ideally in a manner that allows for new important work to surface more easily.

I think what I'm building in peer-review.io actually achieves this. Because I've split out the two functions of review into pre-publication review and post-publication integrity management.

Pre-publication review is entirely about helping authors improve their work. It's rather akin to a Github PR.

Post-publication integrity management happens through voting and responses. More akin to StackExchange. Votes require responses, though responses don't require votes. That part is all public and on going. If you vote on a paper based on reading it, and then later discover its fraudulent in some way, you can come back and change your vote and edit your response. Both review and voting/response stay with the paper in perpetuity.

Also, thanks for the link, I'll give it a read!