Offline Learning for Prompt-Conditioned Biological Interventions

A framework uses vision-language models to judge if archived biological outcomes match natural-language descriptions, enabling training of language-to-intervention mappings without new wet-lab experiments. This approach aims to bridge the gap between linguistic descriptions and complex biological actions through offline dataset utilization.