AIToday
Large Language ModelsAI Safety & AlignmentAI WeeklyPublished: Nov 21, 2026, 01:00 JST

ICLR peer-review overlap jumps from 12.0% to 21.3% as LLM use rises

Send on LINE
ICLR peer-review overlap jumps from 12.0% to 21.3% as LLM use rises

3 Key Points

  1. What happened

    The Nov 19 arXiv paper found ICLR critique overlap rose from 12.0% in 2026 to 21.3% in 2027, with LLM-flagged reviews 4.5 points higher than non-flagged ones.

  2. Why it matters

    The third review at ICLR now surfaces only ~5.46 new problems versus ~6.12 a year earlier, suggesting reviewers are adding less fresh criticism as overlap grows.

WHO IT HITSConference organizers and journal editors who rely on peer review to filter research quality may find their reviewer pool producing less varied feedback, as the same rising overlap pattern shows up across NeurIPS, ICML, F1000 and PLOS.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The arXiv paper, posted Nov 19, quantifies a worry that has circulated quietly since large language models became common writing aids: that peer reviews are converging on similar language. The data point most likely to stick is the 12.0% to 21.3% jump in critique overlap at ICLR between 2026 and 2027, alongside the 4.5-point gap between reviews flagged as LLM-influenced and those not.

The decline in fresh criticism is the mechanism the numbers point to. The third review at ICLR now surfaces only ~5.46 new problems versus ~6.12 a year earlier, meaning later reviewers are repeating more of what earlier ones already said. Because the paper reports the same year-over-year pattern at NeurIPS, ICML, F1000 and PLOS, this appears less like an ICLR-specific quirk than a broad shift in how reviews are written — though the paper stops short of proving what is driving it.

FAQ
What is the main finding about ICLR peer reviews?
Critique overlap at ICLR climbed from 12.0% in 2026 to 21.3% in 2027, and reviews flagged as LLM-influenced had 4.5 points higher overlap than those not flagged.
Does the homogenization show up outside ICLR?
Yes. The paper reports similar year-over-year rises in review overlap at NeurIPS, ICML, F1000 and PLOS.
How much unique criticism does the third review add now?
At ICLR, the third review now contributes only ~5.46 new problems, compared with ~6.12 a year earlier.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleFeed launches Marugoto LLMO to help firms get cited in AI answers