This paper, presented by Mikhail et al., looks at the use of Google's Gemini 2.0 AI to recognize abnormalities in extraocular movement (EOM) from clinical videos. This research was conducted by inputting 114 youtube videos of EOM disorders, along with 15 control videos, into Gemini 2.0. These videos were then cut to only display to ocular exams, removing and text or audio so that the AI model could only go off the exam itself. From these videos, Gemini then was trained to evalutate laterality identification, diagnostic accuracy, movement limitations, and processing time. The Gemini model was able to achieve a diagnostic accuracy of 47.7% across the 114 videos of EOM disorders, while also correctly identifiying 14/15 control videos. For laterality, Gemini 2.0 was able to correctly identify laterality in 26.5% of cases, however, for correctly diagnosed cases, this number shot up to 73.1%. Similarly, 30.4% of cases were identified for movement limitations accurately, with 88.5% of correct diagnoses being identified with movement limitations. Gemini 2.0 excels at correctly identifying features in more obvious disorders with information such as movement deficits, however when challenged with a more difficult diagnosis, it often struggled to differentiate, misclassifying disorders when overfixating on certain details. Nystagmas in particular were difficult, missing all 36 videos, possibly due to the camera being unable to capture these movements. Overall Gemini still lacks accuracy to be used in a clinical setting, however, the framework is in place to run further clinical testing to refine to model.