Sakana AI Scientist

What is Sakana AI Scientist?

The AI Scientist is an automated system designed to handle the full process of creating research papers, from start to finish. It uses foundation models to drive scientific discovery independently. Starting from a basic code template, it moves through generating ideas, searching existing literature, planning and running experiments, creating figures, writing the manuscript, and performing automated peer review. It refines its ideas based on feedback and what it has learned from earlier attempts, mirroring how human scientists work.

Who created Sakana AI Scientist?

Sakana AI developed the AI Scientist, known for applying nature-inspired methods to foundation models. The system was built in partnership with the Foerster Lab for AI Research at the University of Oxford, and with Jeff Clune and Cong Lu at the University of British Columbia. It enables large language models to conduct research independently by automating every step of the research lifecycle, with the aim of making research more accessible and accelerating scientific progress.

Who is Sakana AI Scientist for?

• Research Scientists
• Data Analysts
• Biologists
• Chemists
• Physicists
• Computer Scientists
• Medical Researchers
• Academic Writers
• Science Communicators
• Bioinformaticians

How does Sakana AI Scientist work?

1. Idea Generation: You provide a basic code template related to your topic. The AI Scientist brainstorms research directions and searches Semantic Scholar to check that ideas are novel.

2. Experimental Iteration: The tool runs the experiments it proposes and creates visual plots of the results, with detailed notes for each plot to support the write-up.

3. Paper Write-up: It produces a scientific manuscript formatted in LaTeX, automatically finding relevant papers on Semantic Scholar to cite.

4. Automated Paper Reviewing: An LLM-powered reviewer evaluates the generated paper, designed to be as accurate as a human reviewer. Its feedback can be used to improve the current project or guide next steps.

5. Example Papers: The AI Scientist has produced papers on subjects including diffusion modeling, language modeling, and grokking, demonstrating its ability to find new contributions in machine learning.

By automating the whole research process, from idea generation to writing and reviewing, the AI Scientist helps speed up scientific discovery, particularly in machine learning.