Peer Review

Is peer review important? More AI researchers are posting on arXiv as advanced papers and forgetting to go through the peer review part. Some authors are submitting some version of their peer reviwed articles (like conference papers) through the system.

arXiv is becoming an important part of reports. For example, in the The International AI Safety Report, 30% of the references in the Report (393/1366), 44% of Update 1 (73/168), 60% of Update 2 (100/175), are from arXiv, which is not peer reviewed nor does it mention which version number of the paper it is referring to. The 2026 report had 263 arXiv (or sister publications) out of the 885 (30%) that I looked at =- it has a total of 1451 citations.

The premise of peer review is expert feedback for improvement and scientific verifiability. The assumption is once it is published it has gone through the rigor of those steps. Of course peer review has its disadvantages - it takes time and currently is swamped with slop (too much AI-generated content).

Further, since peer review is not valued and still a professional voluntary task this also creates challenges. How can we change this? Love to hear your thoughts on peer review.