Concerns Rise Over Quality of AI Research Amid Prolific Publications

Kevin Zhu, a recent graduate from the University of California, Berkeley, has claimed authorship of 113 academic papers on artificial intelligence (AI) in 2025, with 89 set to be presented at the NeurIPS conference. Zhu, who runs Algoverse, an AI research and mentoring company, has faced scrutiny from the academic community regarding the quality and legitimacy of his work. Critics, including Hany Farid, a professor at Berkeley, have labeled Zhu's output as a “disaster,” suggesting that it reflects a troubling trend of low-quality research in the rapidly expanding field of AI.

The Guardian

The AI research landscape is characterized by a lack of rigorous peer-review processes compared to traditional scientific fields. Major conferences like NeurIPS have seen a dramatic increase in submissions, with 21,575 papers submitted in 2025, up from under 10,000 in 2020. This surge has led to concerns about the quality of submissions, with reviewers noting a decline in the average scores awarded to papers. Jeffrey Walling, an associate professor at Virginia Tech, highlighted that the review process at NeurIPS is often expedited, compromising the thoroughness typically expected in scientific evaluations.

The Guardian

The pressure to publish has intensified among students and academics, with many feeling compelled to produce a high volume of work to remain competitive. Farid noted that some students resort to “vibe coding” to inflate their publication counts. This phenomenon has raised alarms about the integrity of AI research, as the focus shifts from quality to quantity. Zhu's Algoverse promotes the idea that publications can enhance college applications, further incentivizing students to pursue numerous submissions, regardless of their merit.

The Guardian

Farid has expressed skepticism about the justification for Zhu's extensive authorship, arguing that it is implausible for one individual to meaningfully contribute to over 100 papers. The issue extends beyond individual cases; the use of AI in reviewing submissions has led to concerns about the reliability of feedback and the potential for inaccuracies in citations. A recent position paper from South Korean computer scientists addressed the challenges posed by the influx of submissions and the declining quality of reviews, underscoring the urgency of finding solutions.

The Guardian

The implications of this trend are significant, as the flood of low-quality research complicates the understanding of advancements in AI. Farid warned that the current state of the literature makes it nearly impossible for the average reader to discern credible findings from noise. He advised students to prioritize thoughtful research over mere publication volume, cautioning that the current environment favors those willing to produce subpar work.

The Guardian

NeurIPS organizers acknowledged the strain on their review system due to the rapid growth of AI as a field. They noted that the increasing number of submissions has necessitated a reliance on less experienced reviewers, which may further impact the quality of the conference's output. As the AI research community grapples with these challenges, the need for a reevaluation of publication practices and review standards has become increasingly apparent.

The Guardian

• “I’m fairly convinced that the whole thing, top to bottom, is just vibe coding,” — Hany Farid, Professor of Computer Science, UC Berkeley

The Guardian

• “The reality is that often times conference referees must review dozens of papers in a short period of time, and there is usually little to no revision,” — Jeffrey Walling, Associate Professor, Virginia Tech

The Guardian

• “You have no chance, no chance as an average reader to try to understand what is going on in the scientific literature. Your signal-to-noise ratio is basically one. I can barely go to these conferences and figure out what the hell is going on.” — Hany Farid, Professor of Computer Science, UC Berkeley

The Guardian