Learning to Discover at Test Time
What this work does
Learns during a single target problem using a tail-emphasizing entropic reward objective and state reuse that favors promising candidates while preserving exploration.
Connection to discovery
Distinguishes producing one exceptional verified artifact from maximizing average answer quality, and makes test-instance-specific policy learning explicit.
Evidence
The body states continuous evaluators, validity checks, maximum-child reuse statistics and matched-sampling best-of-N comparisons across mathematics, kernels, algorithms and analysis.
Scope and limitations
Known executable evaluators and an existing candidate language define the discovery boundary. Better best-found score on one target does not establish general discovery transfer or autonomous formulation of scientific validity criteria.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.