Re:Image GPT
by markchen90 on Wednesday June 17, 2020 @08:21PM
(#23558428)
Author here. You're absolutely right that "understanding" is a fuzzy word. As you pointed out, part of reason we hold this belief is that the model can generate diverse samples and successfully complete out-of-distribution inputs. But the other part is that the model learns (without labels) useful features for classifying objects. Would be very interesting to test it on a broader set of datasets which measure other 2D image characteristics.
[ Parent
| on HN ]
There are no comments on this comment yet.