AI research is growing 7× faster than 9 years ago. 191 papers per day. The knowledge is compounding faster than anyone can read. The field is eating itself.
70,000+AI papers on ArXiv in 2024
7×Growth since 2015 (10,000/yr then)
191New AI papers per day (2024 avg)
48hrsHalf-life of relevance for some subfields
Choose your depth. The data doesn't change — just the explanation.
Scientists share their AI research on a website called ArXiv before it's officially published. In 2015, about 10,000 AI papers were posted there per year. Now it's over 70,000 — that's about 191 new papers every single day. Even the world's best AI researchers can only read a fraction of what's being published. The field is moving faster than humans can track.
ArXiv is where AI researchers post pre-prints — research papers before peer review. The AI category (cs.AI, cs.LG, cs.CL, stat.ML) grew from ~10K papers/year in 2015 to 70K+ in 2024. This creates an information overload problem even for experts. Most researchers now use ML-powered tools (Semantic Scholar, Connected Papers) to navigate the literature. The rapid publication pace also means more low-quality work gets attention, creating a signal-to-noise problem in an already fast-moving field.
ArXiv monthly statistics (arxiv.org/stats/monthly_submissions) show cs.AI + cs.LG + cs.CL + stat.ML combined reaching ~5,800/month in 2024. The acceleration correlates with ChatGPT launch (Dec 2022) — submissions jumped 40% in first 3 months post-launch. Quality distribution: Pareto-heavy, with top ~5% of papers (later published in NeurIPS/ICML/ICLR/ACL) accounting for ~80% of citations. AI paper submissions now represent ~35% of all ArXiv submissions across all domains — up from ~15% in 2015.
Data: arxiv.org/stats/monthly_submissions (downloadable CSV). Categories: cs.AI, cs.LG, cs.CL, cs.CV, cs.RO, stat.ML. Semantic Scholar API (api.semanticscholar.org) for citation analysis. Connected Papers (connectedpapers.com) for literature mapping. Epoch AI database (epochai.org) tracks notable papers by compute used. NeurIPS/ICML/ICLR acceptance rates have held roughly constant (~15-25%) while submissions grow — creating an expanding rejected-but-published-on-ArXiv pool. Note: not all cs.* ArXiv papers are AI; filter by abstract keyword for precision.
The Research Explosion
ArXiv AI submissions have grown nearly exponentially. The ChatGPT moment in December 2022 triggered an additional surge — researchers flooded the preprint server with transformer and LLM research.
ArXiv AI Submissions Per Year (cs.AI + cs.LG + cs.CL + stat.ML)
January 2022 through December 2024. ChatGPT launch marked.
ArXiv AI as % of All ArXiv Submissions
AI research is crowding out other fields on ArXiv.
The Comprehension Gap
At 191 papers per day, a researcher who read one paper per hour, 8 hours per day, 7 days a week would still only read 2,912 papers per year — 4% of total output. The field produces knowledge faster than any human can consume it, creating pressure to use AI to read AI research.