Skip to content

Conversation

@Hussain0327
Copy link

@Hussain0327 Hussain0327 commented Nov 27, 2025

I’ve been following the work coming out of Sakana AI for a while, and The AI Scientist feels like a major step forward for automated research. As a student deeply interested in both AI and Japan’s growing innovation ecosystem, I’m fascinated by how intelligent systems can learn, adapt, and ultimately contribute to solving real problems for Japan and the world.

Related issue

Closes #200.

What this PR adds

This PR introduces a production-ready Google Colab notebook that lets users run The AI Scientist without any local setup, addressing the request in #200.

Changes:

  • Add notebooks/ai_scientist_2d_diffusion.ipynb with full pipeline support on Colab
  • Add pinned dependencies under requirements/ for reproducibility
  • Update README.md with a Colab quick start section so new users can get started quickly

Notebook features

The Colab notebook is designed to feel as close as possible to "open and run":

  • No local installation required, only a browser and an API key
  • Supports OpenAI, Anthropic, and DeepSeek APIs
  • Includes a simple cost breakdown and estimates for typical runs
  • Runs the full pipeline: idea generation, experiment updates via Aider, paper writing, and PDF export
  • Automatically finds and downloads the generated PDF at the end of the run

Testing

Tested on Google Colab with a free T4 GPU:

  • Verified full setup and installation flow
  • Confirmed idea generation works by calling the function API directly instead of the CLI wrapper
  • Fixed handling of the novel field when running in --skip-novelty-check mode
  • Verified that Aider modifications apply correctly and that the pipeline completes without errors
  • Confirmed that a complete research paper is generated and downloadable from Colab

Under typical settings the full pipeline runs in about 30-45 minutes on a free T4 GPU, with an estimated cost in the 3-10 USD range depending on model usage.

This PR adds a production-ready Google Colab notebook that enables users
to run The AI Scientist without local setup, addressing issue SakanaAI#200.

Changes:
- Add notebooks/ai_scientist_2d_diffusion.ipynb with full pipeline support
- Add pinned dependencies in requirements/ for reproducibility
- Update README.md with Colab quick start section

The notebook has been tested end-to-end on Google Colab and successfully
generates complete research papers in 30-45 minutes for $3-10 using the
free T4 GPU.

Key features:
- Zero local installation required
- Supports OpenAI, Anthropic, and DeepSeek APIs
- Includes cost breakdown and estimates
- Runs full pipeline: idea generation, experiments via Aider, paper writing
- Auto-downloads generated PDF

Testing notes:
- Verified on Google Colab with free T4 GPU
- Fixed idea generation to call function directly vs CLI
- Fixed novel field handling for skip-novelty-check mode
- Confirmed full pipeline execution including Aider code modifications

This makes The AI Scientist accessible to researchers and developers
without access to local GPU infrastructure.
Users should make a copy before running to avoid losing progress
Keep Sakana AI's original emojis in header and links
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Feature Request: Enhancing Accessibility and Efficiency by Hosting on Google Colab

1 participant