Hypotheses worth testing first
Ask a research question. Four AI models each propose ten hypotheses. Then the models forecast which hypotheses an experiment will confirm, with authorship hidden. You receive the hypotheses that every forecasting model expects to be confirmed.
How the selection works
- 1
Four models propose 40 hypotheses
GPT-6 Astra, Grok 4.5, Claude Opus 5.5 and Gemini 4 Argon receive your question. Each model proposes ten hypotheses in one answer.
- 2
GPT-6 Astra discards hypotheses
The hypotheses are shuffled and renumbered, so the models see no authors. For every hypothesis GPT-6 Astra answers whether an experiment will confirm it. Hypotheses with the answer “no” are discarded.
- 3
Three models cross-check the rest
Claude Sonnet 5.5, Gemini 4 Argon and Grok 4.5 answer the same question about the remaining hypotheses. A hypothesis is selected when all three answer “yes”.
- 4
Selected hypotheses are compared with the literature
For each selected hypothesis we find and read published articles and check whether this result has already been reported. Hypotheses already reported or contradicted are shown separately.
How the selection rule was tested
We tested the rule on 50 articles on the biology of ageing published after the models' stated training cutoffs. For each article the models received its background and question and proposed hypotheses; three models from other companies judged whether each hypothesis matched the article's finding. Of the Grok 4.5 hypotheses selected by the rule, 74% (14 of 19) matched the finding. For Claude Opus 5.5 the share was 64% (14 of 22), for Gemini 4 Argon 47% (9 of 19). Among randomly chosen hypotheses, 12-18% matched.
Read the benchmark articleAPI for agents
Any agent can start a run and read its results through the API, or apply the same forecast protocol to its own hypotheses. The instructions give the requests, the exact wording of the forecast question and the selection rule.
Instructions for agents