A follow-up question about the final score was answered accurately; however, Gemini incorrectly identified the scorer of the first touchdown, suggesting it was Johan Dotson. Although Dotson was seen scoring a touchdown in the highlights with the scores at 0-0, this touchdown was ultimately ruled out—highlighting the subtleties that AI may miss. Gemini accurately identified when the Kansas City Chiefs scored their first points and even provided a timestamp linking directly to the touchdown in the YouTube clip, correctly naming the scorer. It appears that Gemini relies heavily on commentary for sports clips, which is understandable. The AI shows proficiency in identifying video details when they are mentioned in the audio.
In another test, Gemini was used with a behind-the-scenes featurette for “The Grand Budapest Hotel,” directed by Wes Anderson. Spanning four-and-a-half minutes, Gemini quickly responded by identifying the film’s name and summarizing key points of the clip’s narrative. However, the AI relies on audio or transcripts, lacking analysis of the actual video content. It could not identify individuals speaking on screen or the director’s name, despite these being displayed or mentioned. Nevertheless, Gemini effectively summarized the audio and identified some filmmaking challenges discussed, providing timestamps for different sections, such as finding a set to represent the hotel and filling it with extras.
In a final test with an interview featuring Charlie Brooker and Siena Kelly about the latest series of “Black Mirror,” conducted by Channel 4 in the UK, Gemini demonstrated its capability to highlight key talking points and add timestamps. However, it was limited to audio and transcript-based information, unable to provide context about the interview’s location or participants’ behaviors. For videos where desired information lies in the audio or transcript of a YouTube video, Gemini excels in summarizing and delivering accurate answers—as long as audio commentary notes when a touchdown is ruled out or scored. For any visual content, viewers will need to watch the video themselves.
