Academic Concerns Arise Over Prolific AI Research Output and Quality
Kevin Zhu, a recent computer science graduate from UC Berkeley, claims to have authored 113 AI research papers in 2025, with 89 slated for presentation at a top AI and machine learning conference. Zhu runs a company called Algoverse alongside high school co-authors and mentors teams through a 12-week online program costing $3,325 that aids in methodology, study design, and drafting. His research topics range from locating nomadic pastoralists in sub-Saharan Africa to skin-lesion evaluation and Indonesian dialect translation. Zhu states his LinkedIn profile reflects over 100 top conference papers in the past year, citing recognition from major tech companies and universities, and involving co-authors who are often mentees.
However, this prolific output has sparked controversy within the academic community. Hany Farid, a UC Berkeley computer science professor, criticizes Zhu's approach as "vibe coding" and describes the trend as a disaster for the field, highlighting pressure that prioritizes quantity over quality. The review process for AI research is less stringent compared to disciplines like chemistry and biology. For example, NeurIPS received 21,575 papers in 2025, while ICLR 2026 submissions are expected to reach about 20,000, marking a 70% increase from the previous year. Organizers note most submissions are workshop papers subject to lighter review, though some experts argue the sheer volume of papers by single authors does not fully explain the situation.
Moreover, an influx of research papers posted on arXiv by tech companies and AI safety groups has made it difficult to discern the true state of the AI field, exacerbating signal-to-noise challenges. In May 2025, a position paper by three South Korean computer scientists addressing the surge in submissions won an ICML 2025 award. Farid has advised students to avoid pursuing AI research due to concerns about the current research frenzy and overall quality decline.
