OpenAI introduced BioBasicsBench, a 100-task test built with Ligo to check whether AI agents can interpret wet-lab data, spot errors and tell when evidence is too weak to support a conclusion. di.gg/s/nrvw3ty5