Why it matters
Experiment selection is a distinct research-agent capability with a clear budget use case. The current evidence is an author evaluation with unresolved transfer limits.
For AI researchers, Experiment-platform builders.
Sources & further reading
Paper · paper
Research World Models ↗
October 8 abstract describes predicting experiment outcomes from historical records under limited research budgets.
Research World Models authors · page checked 2026-10-16
Coverage note: The sources were read on October 16. Independent execution and a relevant YouTube/X comparison remain unavailable.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
Research World Models explores predicting the results of proposed AI experiments from records of earlier experiments. The October 8 paper reports tests over multiple research environments and limited budgets. Researchers may use this idea to prioritize expensive runs, while real experiments still determine whether a prediction holds. The public abstract was read October 16. Training records, implementation and the reported gains were not independently reproduced.