Plan the research
A researcher defines a question, checks the existing literature and uses their own AI access to draft tasks, expected outputs and validation rules.
Research takes people. AI can help with the work.
Help researchers turn published evidence into something others can check and use. Explore three proposed projects: clearer blood-pressure measurements, cooler homes, and evidence on metformin as an adjunct to TB treatment.
Meet the first projects ↓Recruiting researchers, reviewers and future contributors.
Project planning is open. Project-specific web apps support manual contributions.
Health / measurement science
When do shorter home blood-pressure monitoring schedules agree with longer ones?
No patient data. No personal medical recommendations.
The idea behind Tokens4Good
Find the comparison in this paper. Check the units. Verify the source. Notice where two studies disagree.
AI can help with this work, but a pile of generated answers is not a research result. A researcher needs to define the method, check the output and make sense of the evidence.
We are building a way for people with AI access to help with well-defined parts of that work, one approved batch at a time.
A researcher's question. A shared effort.
A researcher defines a question, checks the existing literature and uses their own AI access to draft tasks, expected outputs and validation rules.
The lead reviews a bounded group of tasks. Nothing enters the queue simply because an AI suggested it. The first batch tests the method as well as the runner.
Open the project’s dedicated web app to claim approved tasks and return evidence. AI-plan integration is still a separate, unverified step; manual participation does not connect a provider account.
Researchers review source-linked results, reconcile disagreements and combine accepted work. Follow-up questions need another approved batch.
Researcher approval and contributor permission are separate. We do not run research on the strength of a website signup.
Three starting questions
All projects are in planning. We are looking for qualified leads to challenge these drafts before any batch runs.
Health / measurement science
Planning / recruiting a leadWhen do shorter home blood-pressure monitoring schedules agree with longer ones?
Map how the number and timing of home readings affect agreement and repeatability in published studies. No new patient data. No personal medical advice.
Read the research plan ↗Climate / building science
Planning / recruiting a leadUnder what conditions do reflective roofs reduce indoor heat in homes without air conditioning?
Compare measured indoor-air outcomes across roof types, weather and ventilation conditions. Find where the evidence is convincing and where it is thin.
Read the research plan ↗Health / open drug repurposing
Planning / recruiting a leadWhat does human evidence show about adding metformin to tuberculosis treatment?
Map clinical outcomes, adverse events and drug interactions in published studies of adjunctive metformin for pulmonary TB. Separate trial evidence from observational associations. No treatment recommendations.
Read the research plan ↗A useful result needs a method
Each task has an input and a check.One permitted paper, one comparison, one structured record. Missing information stays missing. Every extracted claim needs a source.
Accepted work becomes an inspectable artifact.The aim is an evidence table, a documented method and an account of limitations, not a polished AI summary presented as science.
Projects can change direction.New evidence can create new questions. The project lead decides what goes into the next batch.
Before you join
We are opening recruitment,
not announcing completed research.
Read the project plans and register interest as a contributor, researcher or reviewer. The starting proposals are seeking research leads. Researchers can create their own project pages and task plans; contributors can use each project’s web app when approved tasks are available. AI execution is not enabled.
Eligible AI requests made through an open-source app using your own subscription allowance. You would not transfer a token balance, share a password, or donate your computer's processor. Research requests would count toward limits you also use elsewhere.
A named project lead defines the method. AI can help draft small tasks, but a researcher must approve each batch before contributors can take it on. You choose a project and separately authorize your own participation.
A result can suggest a gap, disagreement or follow-up. Those ideas become a draft batch, not automatic permission to run more work. The researcher reviews it before anything enters the queue.
The pilot protocols call for independent extractions, verification of source passages and quantitative fields, and researcher adjudication. We would keep negative findings, missing data and disagreements. A generated answer is not a validated finding.
No. They examine published aggregate research, not your readings or treatment. They do not recruit patients, make diagnoses or recommend medicines, doses or monitoring schedules. Qualified clinical and methods reviewers must approve each protocol before execution.
An eligible open-source app using the documented ChatGPT plan-usage flow is the intended first runner. Authorization, usage controls and end-to-end execution still need testing. Website registration does not connect an OpenAI account or authorize inference. We are independent, not an announced OpenAI partner.
The planned app keeps AI credentials on your device, shows when your plan is being used and offers pause/stop controls. The pilots start with public or explicitly permitted material, not personal files or patient records. The interest form only collects recruitment details described in our privacy notice.
These questions have existing literature. The aim is a reproducible, source-linked synthesis, not a promise of discovery. Before execution, the lead must check recent reviews, refine the gap and decide whether a new project adds value.
We intend to preserve contributor attribution and publish permitted evidence tables, methods and review history. Each project needs an explicit data and output license before work begins. Contribution credit is not a guarantee of paper authorship.
Integration background: OpenAI's open-source app guide ↗. Provider eligibility and limits apply.
The next step is people
Bring subject knowledge, a skeptical eye, or an AI allowance you would like to put to work when the app is ready.
Research leadsRefine the question. Own the protocol. Interpret the findings.
ReviewersChallenge the method. Check the sources. Resolve disagreements.
ContributorsChoose a project. Keep control of your access. Help with a bounded task.
No payment, patient data or AI account connection. You choose the project and how you would like to help.