A real research problem
A task from your experiments, analysis, code, recordings, protocols, specialist software, or a whole research project.
Contribute to Science’s Last Exam
Bring your hardest problem. Let’s collaborate
- Join the co-author list on the SLE paper.
- Shape how scientific agents are evaluated.
- Collaborate with researchers across disciplines.
Submission guide
A scientific challenge
Requires scientific knowledge and reasoning.
A verifiable result
Include a reference result, executable checks, or a scoring rubric.
- 01
Goal
The question, starting state, and constraints.
- 02
Inputs
Data, code, recordings, or documents the agent can use.
- 03
Environment
Tools, versions, licenses, and runtime requirements.
- 04
Outputs
The files, results, or decisions the agent must deliver.
- 05
Evaluation
Reference results, tests, and tolerances—kept separate from agent inputs.
Files: Upload up to 50 MB per file. For larger files, add a Google Drive / Dropbox link and give the review team access.
Review: Complete the five items, then follow review feedback in My submissions.
An AItonomy initiative. Researchers write and review tasks in a personal capacity.
- Institutions
- 34
- Countries · continents
- 5 · 4
- Domain experts joined
- 43
- Expert recruitment target
- 200+
Contact
Research contactsly1988@princeton.eduwuyc@stanford.edu