pull down to refresh

Another point that I haven't seen brought up much is that the AI will never stop nitpicking. If you ask it to review a paper, then revise it to address every criticism, then ask AI to review it again, it will come up with a whole new set of criticisms.

Sometimes the critiques even contradict themselves, especially when it comes down to judgment calls.

I really think the traditional model of publishing is due for a reckoning.

I hope for a future of collaborative research via open source models

Maybe we need to bring back the ralphwiggin skill for reviews so that we get all the findings at once instead of dripping them in, because I think the most annoying part, especially with recent Claude "improvements", is that it always leaves things out. I rarely get a list with more than 12-15 findings, then after addressing these, another 12-15 findings.

reply

Not sure how it is for code review, but for research review you sometimes get this cycle

  1. This finding is interesting, you are selling it too short, you should emphasize it more!

--> go back and emphasize that result

  1. You are overselling this result. Your interpretation is not justified by the evidence. Tone down your claims!
reply

Those typically occur in the same round of reviews by different reviewers.

I have yet to have the same reviewer give me that feedback in successive rounds.

reply

I've had it on a bug a week and a half ago. The first review told me "this is incomplete so add xyz", the second (after implementing it) said "remove xyz because it goes against spec".

reply

oh. Yeah that's different for me. I expressly ask for evidence, and often even don't ask for solutions. There's still a similar symptom, but just like I don't argue with bots, I really do not give 2 shits about the opinion of a bot. Execute the task, and hup hup.

reply
43 sats \ 0 replies \ @ek 2 Sep

You might be interested in Fulcrum

reply
a future of collaborative research via open source models

That would be great. I'd prefer that so much to the anonymous review process.

If someone thinks they can make a meaningful improvement to my work, I'd rather bring them on as a co-author who I can actually interact with.

And if they think my work is utter shit, I'd rather they do the work to publish a rebuttal themselves, rather than have to seek their approval to publish my own work.

reply