I've been around the block a few times. When deep learning first became hot, many older colleagues bemoaned it as just tinkering + chain rule, and not intellectually satisfying. Then came SSL, equivariance, VAEs, GANs, neural ODEs, transformers, diffusion, etc. The richness was staggering.

🧵👇

Hard to say anything conclusive right now. When we investigated low quality reviews, the review was more often in the control group (anecdotal impression).

But there are whole categories of issues that the feedback agent wasn't looking for. I think actively looking for them would have helped.

We evaluated this while planning ICLR 25, but decided against it because it disproportionately harms students. Senior researchers are the ones submitting multiple papers.

We sent a reminder email to reviewers today - you could also reply to your response in a few days to ping the reviewers. That will trigger an email reply to the reviewers.