Measuring how AI agents handle everyday wet-lab science

OpenAI has released BioBasicsBench, a benchmark containing 100 tasks, to evaluate the ability of AI agents to interpret raw experimental data and handle daily wet experimental science.