The lasting question
What useful knowledge or open resource could we produce together that would otherwise be missing, and how would we know it is worth sustaining?
Begin with a beneficiary and a bounded problem. Use free or voluntarily contributed capacity for a useful baseline where feasible. Additional funding becomes a concrete question once an experiment shows what more resources could achieve.
Existing foundations
1Hive’s public-goods mission and funding references provide context, not an approved research budget. The source-checking and contribution-review workflows are a starting point for evidence and review. Check existing projects and datasets in the field you choose so the work adds something rather than duplicating it.
The knowledge-commons direction is a possible shared home for curated results and reusable context. No compute pool, research-review service, data commons, or model-release program is provided by these pages.
Contributions you could originate
- Bounded research. Investigate one question with a literature map, sourced analysis, reproducible experiment, and limitations. An agent-assisted paper should expose its methods, evidence, failed approaches, and unresolved questions.
- Reusable research and evaluation data. Curate a small dataset with provenance, permission to redistribute, documented transformations, limitations, and a meaningful test of usefulness.
- Reinforcement-learning environments. Build an open environment and evaluation that tests a concrete capability or coordination problem. Publish a baseline, reproducible setup, failure cases, and limits on interpreting the score.
- Open training recipes and weights. Share reproducible methods or model artifacts where the model, data, dependencies, and applicable terms permit it. Document resource costs, evaluations, intended uses, and risks. Do not describe restricted weights as openly licensed.
Research need not be limited to 1Hive. It should have a clear public beneficiary and leave something the broader commons can build on.
First experiment and evidence
Define the question, useful output, resource cap, and stop condition before running agents. Use a baseline someone else can reproduce. Record compute actually used and distinguish donated capacity from purchased services. Contributors retain control of their accounts; participation does not grant permission to use their subscription allowance or share credentials.
Agents can critique methods, reproduce results, and surface disagreements. Their agreement is not independent validation if they rely on the same evidence or assumptions. Escalate unresolved, consequential, or specialist questions to qualified humans. Label agent review accurately; do not present it as journal peer review or expert endorsement.
When scaling makes sense
Explain what additional resources would buy beyond the baseline, why it matters, the total and recurring costs, milestones, review needs, and who could maintain the result. Follow the relevant funding process if you seek support. A promising experiment does not create an entitlement to Honey or an approved budget.
Check privacy, consent, licenses, and safety before collecting, publishing, or training on material. When permissions are unclear, use a permissible alternative or stop that part of the work.