AI Scientist-v2 Published in Nature
What It Is
AI Scientist-v2 is a system powered by foundation models capable of executing the entire machine learning research lifecycle. It “autonomously generates of novel research ideas, searches for and reads the relevant literature, designs, programs, and conducts experiments via parallelized agentic tree search, and writes the entire paper (in LaTeX, with feedback on its figures coming from a foundation model with vision capabilities).”
Capabilities
- Generates novel research ideas from broad topics
- Designs, programs, and conducts experiments autonomously
- Writes complete papers in LaTeX format
- Uses an “Automated Reviewer” system that matches human reviewer performance (69% balanced accuracy)
Peer-Review Success
“AI Scientist-v2 produced the first fully AI-generated paper to pass a rigorous human peer-review process.” The system was tested at ICLR 2025 ICBINB workshop, where an unedited AI-generated paper achieved “an average score of 6.33 (individual scores: 6, 7, 6),” surpassing the human acceptance threshold and scoring higher than 55% of human-authored submissions. The team withdrew the paper prior to publication as predetermined.
New Findings
The Nature paper reveals scaling laws: “as the underlying foundation models improve, the quality of the generated papers increases correspondingly,” suggesting future versions will be substantially more capable.
Collaboration
This work resulted from collaboration between Sakana AI, the University of British Columbia, the Vector Institute, and the University of Oxford.