OMNI-EPIC: Open-endedness via Models of human Notions of Interestingness with Environments Programmed in Code
What this work does
Uses language/vision models to generate task descriptions and executable environments, train agents, and retain learned/failed tasks in an evolving archive.
Connection to discovery
A concrete model for generating discovery-training worlds and curricula from a persistent record of competence and failures.
Evidence
Inspected archive-conditioned generation, simulator reset/step/reward construction and visualized task lineages.
Scope and limitations
Open-ended generation remains bounded by the simulator, code APIs and evaluators. Generated-task success is not evidence of transfer to unmodeled real-world science.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.