OpenAI Releases LifeSciBench: Measuring the Capabilities of AI Systems in Real-World Scientific Research Scenarios

Odaily Planet Daily News OpenAI has released a new evaluation benchmark, LifeSciBench, designed to measure the capabilities of AI systems in real-world scientific research scenarios. It is reported that LifeSciBench is based on 750 expert-written tasks, covering 7 types of scientific research workflows and 7 biology fields. The tasks were contributed by 173 researchers with Ph.D. backgrounds and experience in the biotechnology or pharmaceutical industries. This benchmark emphasizes the assessment of complex scientific research capabilities, including evidence synthesis, experimental design, data analysis, scientific reasoning, and scientific communication, rather than simple fact-based questions. Over 79% of the tasks involve multi-step reasoning, with each task requiring an average of about 4 reasoning steps, and includes 1,062 data appendices related to real scientific research (such as papers, charts, sequence data, and structural files).