AI SCIENCE
Someone confiscate the printer
AI research conferences are struggling with a massive rise in paper submissions.
The International Conference on Learning Representations (ICLR), one of AI’s biggest conferences, received around 500 papers a decade ago.
For its 2027 event, that number has climbed to over 40,000.
The rise comes from growing interest in AI, pressure to publish research, and AI tools making papers quicker to write.
But there’s one big problem: conferences rely on researchers to review each other’s work, and there aren't enough qualified people to keep up.
Other organisations are also making changes:
NeurIPS is cracking down on poor-quality reviews and requiring researchers to finish their assigned reviews before seeing feedback on their own work.
AI-written papers are still allowed because AI detectors aren't reliable enough, and researchers haven't agreed on how much AI use is acceptable.
arXiv, a popular research-sharing platform, is limiting authors to two papers per month after receiving over 40,000 submissions in September, nearly double the number from two years ago.
The world's least enjoyable book club
To manage the workload, ICLR and the Association for Computational Linguistics’ Rolling Review (ARR) system are bringing in new rules.
Both are limiting most authors to 20 papers per submission period, while ARR will use a lottery to decide which papers without qualified reviewers get reviewed.
The goal is to cut down on low-quality submissions and give reviewers a more manageable workload.
However, some researchers worry the limits could make things harder for younger academics or fail to solve the bigger problem.
AI is out there writing every student’s thesis. - MV