Ethical disclosures and Gaussian Splatting are on the wane, while the sheer volume of submitted papers represents a new problem for AI to tackle in 2026.
Research trends in AI have seen significant changes recently, particularly in the volume of submissions to platforms like arXiv. As the landscape of computer vision and image synthesis evolves, the sheer number of papers being submitted has raised concerns among researchers about the future of meaningful academic discourse. A report from December 2025 highlights these trends, noting that the volume of AI-related papers has reached a crisis point, further exacerbated by a surge in AI investment.
The rise in submissions, characterized as exponential, has been fueled by the recent investment boom in AI, leading to a flood of new research papers, particularly during the fall conference season. The data shows that computer science submissions have consistently outpaced other categories, drawing attention to the overwhelming influx of research and the consequent difficulties in discerning significant findings amidst a growing “signal-to-noise” ratio.
In response to this deluge, new analytical tools are emerging to evaluate the novelty of research submissions. One standout is the NoveltyRank system, which uses fine-tuned large language models (LLMs) like Qwen3-4B-Instruct-2507 and SciBERT to classify papers based on their novelty. Despite its promise, the methodology has raised concerns among experts, particularly the assumption that conference acceptance equates to novelty and the inherent value of “new” research.
This approach overlooks the complexities of academic research, where many submissions adhere to the “publish or perish” mentality, leading to minimal innovations that may not significantly advance knowledge in the field. Critics argue that AI’s current capabilities lack the long-term contextual understanding necessary to accurately assess the value of research, often resulting in false positives.
As the volume of submissions rises, a notable decline in ethical disclosures has been observed. Historically, ethical statements were a common requirement in biological sciences; however, their presence in AI-related submissions has dramatically decreased throughout 2025. Analysts speculate that this shift may stem from recent deregulation efforts by the U.S. government, which have created an environment perceived as less legally risky for researchers.
The generative video sector has also transformed dramatically this year. Following the launches of Hunyuan Video and WAN series, significant advancements in AI video technologies have emerged, easing previous challenges associated with creating realistic avatars and dynamic video content. The competitive landscape has now been bolstered by numerous companies, including civit.ai and RunPod, which are capitalizing on these advancements with user-friendly platforms.
Despite the technological leaps, many submissions in the generative video category are characterized as incremental, with core challenges still unresolved, such as maintaining identity consistency and improving audio generation. Researchers are now facing a saturation in submissions, complicating the identification of genuine advancements.
Conversely, interest in mesh-based approaches, previously gaining traction, has waned significantly by the latter half of 2025. The rise of diffusion-based generative systems like Veo and Kling appears to have overshadowed mesh-based solutions, which had seen renewed interest last year. Gaussian Splatting, once considered cutting-edge, is now experiencing diminished momentum, as researchers grapple with its high resource demands.
Additionally, a notable decline in submissions related to AI security has been reported, raising questions about the underlying factors contributing to this trend. The Cryptography and Security archive at arXiv has historically been a less favored platform for researchers, often dominated by proprietary private sector work that seldom appears in open-access venues.
Looking ahead to 2026, the research community is grappling with a paradoxical situation. While the Gen AI boom is often likened to the dot-com bubble of the early 2000s, the current landscape presents unique challenges and opportunities. As AI continues to permeate research culture, measures may emerge to mitigate the overwhelming volume of AI-generated submissions, including potential bans or checks on AI-influenced papers, echoing recent actions by arXiv concerning review papers.
As the field evolves, the interplay between funding, research integrity, and the quest for meaningful contributions will define the trajectory of AI research in the coming year.
See also
AI Systems Projected to Consume Up to 765 Billion Litres of Water Annually, Surpassing Bottled Water Use
AI Research in Cardiovascular Disease Surges 400% Since 2018, Reveals Key Trends and Insights
IIT Delhi’s AILA AI Runs Complex Lab Experiments Independently, Reducing Time by 90%
AI Study Maps 100 Years of Aging Research, Reveals Key Trends and Gaps
