When Masters Fail: GPTZero Detects NeurIPS AI Paper Hallucinations in Global Tech Research
GPTZero detected 100 hallucinated citations in 51 NeurIPS papers. Despite being a small percentage, it highlights the 'submission tsunami' and the irony of AI experts falling for LLM errors.
It's the ultimate irony in the AI world. Even at the industry's most prestigious gathering, NeurIPS, AI-generated slop has managed to sneak into peer-reviewed research papers through the back door of fake citations.
GPTZero detects NeurIPS AI paper hallucinations across 51 studies
AI detection startup GPTZero scanned all 4,841 papers accepted by the conference held last month. According to TechCrunch, the firm identified 100 hallucinated citations across 51 papers. These were citations to papers that simply do not exist, a classic symptom of LLM hallucinations.
Statistically, the impact seems minor. With only 1.1% of papers affected, the conference organizers told Fortune that the core research findings aren't necessarily invalidated. However, the fact that the world's leading AI experts, whose reputations depend on accuracy, couldn't catch these errors in their own work raises serious questions about the future of scientific publishing.
The Submission Tsunami Straining Peer Review
GPTZero's report points to a "submission tsunami" that's straining the peer review pipeline to its breaking point. Citations act as a form of academic currency, and when AI fabricates them, it devalues the metric for everyone. Reviewers, overwhelmed by the volume, are finding it nearly impossible to fact-check every single reference.
Authors
Related Articles
Samsung Electronics' union branch revealed that 84% of survey respondents oppose the government's roughly $290 billion (₩400 trillion) semiconductor megaproject in Gwangju. Government, management, the union and shareholders are all reading the same national project in sharply different ways.
Nvidia shipped roughly a billion RISC-V cores in 2024, then announced it would run CUDA on the open standard. We break down how royalty-free instruction sets and open software stacks are trying to route around CUDA's lock-in. Part 2 of the Semiconductor Sovereignty series.
US AI-chip export controls split into three layers in the first half of 2026 — January easing, a May crackdown on circumvention, and a pending bill. Nvidia erased China from its guidance and still posted a record $81.6 billion quarter. A look at the export policy that both shields and cages it.
AMD's MI325X matches or beats Nvidia on memory and bandwidth — yet Nvidia's 86-92% share holds. The real moat is CUDA, 20 years in the making. Part 1 of 4.
Samsung Electronics' union branch revealed that 84% of survey respondents oppose the government's roughly $290 billion (₩400 trillion) semiconductor megaproject in Gwangju. Government, management, the union and shareholders are all reading the same national project in sharply different ways.
Nvidia shipped roughly a billion RISC-V cores in 2024, then announced it would run CUDA on the open standard. We break down how royalty-free instruction sets and open software stacks are trying to route around CUDA's lock-in. Part 2 of the Semiconductor Sovereignty series.
US AI-chip export controls split into three layers in the first half of 2026 — January easing, a May crackdown on circumvention, and a pending bill. Nvidia erased China from its guidance and still posted a record $81.6 billion quarter. A look at the export policy that both shields and cages it.
AMD's MI325X matches or beats Nvidia on memory and bandwidth — yet Nvidia's 86-92% share holds. The real moat is CUDA, 20 years in the making. Part 1 of 4.
Thoughts
Share your thoughts on this article
Sign in to join the conversation