Inside GeneBench-Pro: Why Testing AI on Biology Is Really a Test of Judgment
Ask a biologist what makes their job hard, and the answer is rarely “not knowing enough facts.” It’s the moment when a dataset looks slightly off — a sample that doesn’t match its supposed ancestry, a signal that could be biology or could be a instrument quirk — and someone has to decide what that means and what to do next. OpenAI’s new benchmark, GeneBench-Pro, is built entirely around that moment. It isn’t really asking “does the AI know genomics?” It’s asking whether an AI agent can make the same kind of judgment call a careful scientist makes when the data won’t simply hand over an answer.