> It's notoriously awful at extrapolating, ie generalizing to anything without support in the training data. Scientists are also pretty lousy at making new dis

> It's notoriously awful at extrapolating, ie generalizing to anything without support in the training data. Scientists are also pretty lousy at making new discoveries without labs. They just need training data.