The world of artificial intelligence research is facing a severe crisis of quality, with academics warning of a deluge of substandard 'slop' overwhelming prestigious conferences. The issue has been thrown into sharp relief by the case of a single author, Kevin Zhu, who claims to have authored or co-authored a staggering 113 academic papers on AI this year alone.
The Prolific Author and the 'Disaster' Papers
Kevin Zhu, a recent computer science graduate from the University of California, Berkeley, now runs Algoverse, a company that mentors high school students in AI research. An astonishing 89 of his papers are slated for presentation this week at NeurIPS, one of the globe's premier AI and machine learning conferences. His published work covers diverse topics, from using AI to locate nomadic communities in Africa to evaluating skin lesions and translating Indonesian dialects.
However, his extraordinary output has raised profound alarm. Professor Hany Farid of Berkeley labelled Zhu's papers a "disaster", suggesting the work represented little more than "vibe coding" – a term for using AI tools to generate software and research with minimal human oversight. In response to queries, Zhu stated he had "supervised" 131 papers as "team endeavours" through Algoverse, which charges students over $3,300 for a 12-week mentoring programme that includes help with conference submissions.
A System Overwhelmed by Volume
The case points to a much larger systemic failure. Top AI conferences are being inundated with submissions. NeurIPS received 21,575 papers this year, more than double the number in 2020. Similarly, the International Conference on Learning Representations (ICLR) reported a 70% surge in submissions. This flood is compromising review quality.
Jeffrey Walling, an associate professor at Virginia Tech, explained that the conference peer-review process is far quicker and less rigorous than in traditional sciences. Reviewers, often PhD students, must assess dozens of papers in short timeframes, with little opportunity for revision. The Chinese tech blog 36Kr noted that average reviewer scores at ICLR have declined, with complaints about poor quality and suspected AI-generated content.
"The reality is that academics are rewarded for publication volume more than quality," Walling said. "Everyone loves the myth of super productivity."
The Human Cost and the 'Frenzy'
Behind the statistics is intense pressure on students and early-career researchers to amass publications. Professor Farid revealed he has even counselled students against entering AI research due to the "frenzy" and the volume of low-quality work. "It's just a mess. You can't keep up, you can't publish, you can't do good work, you can't be thoughtful," he lamented.
This environment creates perverse incentives. Farid noted that some students resort to "vibe coding" to boost their publication counts. Meanwhile, organisations like Algoverse explicitly market their programmes as a way to strengthen college applications and résumés through prestigious conference admissions.
The crisis has grown so acute that finding a solution is now a subject of academic papers itself. A May 2025 position paper by South Korean computer scientists, proposing solutions to the submission surge and review quality crisis, won an award at the International Conference on Machine Learning.
Furthermore, the problem extends beyond conferences. Major tech firms and AI safety groups now routinely publish unreviewed work on the arXiv preprint server, flooding the ecosystem with content presented as science but lacking any formal scrutiny.
The ultimate cost, according to Farid, is a collapse in clarity. For journalists, the public, and even experts, distinguishing genuine advancement from noise is nearly impossible. "Your signal-to-noise ratio is basically one," he said. "I can barely go to these conferences and figure out what the hell is going on." The field of AI, he warns, is at risk of drowning in its own output.
