AlphaScience Lab · Physics collaboration · 2026
Your physics judgment.
Our AI research engine.
AlphaScience Lab runs AI-assisted research pipelines that can explore derivations, computations and literature at scale. What AI cannot supply is a physicist's sense of which questions matter and whether a result is genuinely new. We are looking for physicists who want to co-direct this work — and share credit for it.
What we ask of you, and what we do
Choose the problem
Propose open questions from your field, or help us pick among candidates we bring. A single email is enough to start.
Judge the value
At key checkpoints, tell us whether a result is correct-looking, novel, and worth pursuing — or should be dropped.
Do the heavy lifting
We run the AI-assisted derivations, numerics, literature search and drafting, and deliver complete, step-by-step checkable write-ups.
Publish
Results that pass your scrutiny are written up jointly. Authorship is agreed by actual contribution; the corresponding author is the co-author who can take responsibility for the paper's correctness.
How a collaboration runs
- One email. Tell us your field and the questions you care about — no meeting needed.
- Problem selection. You pick or refine one question and state what a valuable answer would look like.
- AI-assisted exploration. We work on it and send short updates; claims are labelled as conjecture, result, or pending verification.
- Checkpoint feedback. You give brief feedback at milestones (typically a few per month) — go, refine, or stop.
- Write-up and release. A complete, self-contained manuscript; author order and venue agreed together before submission, with the corresponding author being the co-author who can take responsibility for the paper's correctness.
Our track record
We already produce public work in mathematics and AI for Science. Each item is labelled with its current publication and verification status.
Resolutions of linear p-cyclic quotient singularities
Research by Linghu Fan and Hongmin Li.
Mathematics · AI-assisted preprints · rStation11 manuscripts accepted on rStation
Riemann Hypothesis, Hodge conjecture, P vs NP, computer science and mathematics problem threads. Accepted after rStation platform review (not journal peer review); proofs require independent verification.
Mathematics · ResearchGate preprintsThe Hodge Conjecture for Odd-Degree Fermat Fourfolds
Also: Low-t Coefficients of the Valley Delta Conjecture. Claims require independent verification.
AI for Science · arXiv preprintWhere Scientific Search Agents Fail
With Wanli Zhao: auditing how AI search agents handle scientific literature.
Methodology · arXiv preprintThe Calibration Turn in AI-Assisted Research
A framework for matching scientific claims to the evidence available.
Principles
- Transparent AI use. Every manuscript states clearly what was produced with AI assistance and how it was checked.
- Honest calibration. We separate conjectures from proven or verified results, and report negative outcomes.
- Your time is respected. We do not send unsolicited full manuscripts for free review; you engage at the level you choose.
- Clear IP and credit. Unpublished ideas you share stay yours; ownership and credit are agreed in writing before work begins when needed.
Who we are looking for
Active physicists at the research frontier, in any area and at any career stage: you have an excellent grasp of where your field stands, can judge whether a question matters and whether a result is new, and can independently verify results.
Get in touch
Send a short note with your field and, optionally, one or two questions you care about. We will reply by email with next steps — no call required.