The AI Scientist First Original Research Papers

Are we on our way to achieving AGI?

Automated AI research is the critical step towards the “intelligence explosion”, which will lead to “superintelligence”, a state where AI will be able to continuously self-improve and expand its own knowledge and capabilities. Think of unlimited compute and millions of AI agents operating 24/7, 7 days a week …

We might be closer than we realize.

Well, Sakana.ai (for Portuguese speakers, no jokes, please), an AI research lab from Japan, has just unveiled The AI Scientist, the first system engineered for fully automated scientific discovery, enabling foundation models to independently conduct research (https://sakana.ai/ai-scientist/).

Essentially, Sakana.ai was able to engineer an AI system capable of researching new knowledge, testing it, peer-reviewing it, publishing the findings and then incorporating that knowledge for further discovery.

“The AI Scientist automates the entire research lifecycle, from generating novel research ideas, writing any necessary code, and executing experiments, to summarizing experimental results, visualizing them, and presenting its findings in a full scientific manuscript”.

“We also introduce an automated peer review process to evaluate generated papers, write feedback, and further improve results. It is capable of evaluating generated papers with near-human accuracy”.

“The automated scientific discovery process is repeated to iteratively develop ideas in an open-ended fashion and add them to a growing archive of knowledge, thus imitating the human scientific community”.

In its first demonstration, The AI Scientist conducted research in a few ML subfields, discovering novel contributions in popular areas like diffusion models, transformers, and grokking. Below are the titles of the original AI-generated papers:

* DualScale Diffusion: Adaptive Feature Balancing for Low-Dimensional Generative Models

* StyleFusion: Adaptive Multi-Style Generation in Character-Level Language Models

* Adaptive Learning Rates for Transformers via Q-Learning

* Unlocking Grokking: A Comparative Study of Weight Initialization Strategies in Transformer Models

Most impressively, each paper was developed at a cost of just $15.

Conclusion

It is happening fast.