Google’s Gemini Fake AI Video: Unveiling the Reality
Google’s Gemini fake AI Video has stirred up some questions. The video suggests that Gemini’s capabilities are in doubt, primarily because it doesn’t provide clear information on how it processes inputs.
Google’s Gemini Duck Video: A Closer Look
Let’s discuss the Google Gemini duck video, which uses still images. In a recent demo, Google showcased parts of the viral duck video, pitting Gemini against GPT-4, a competitor. Google was transparent about the video, titled “Hands-on with Gemini: Interacting with multimodal AI,” stating that it was edited to speed up the outputs, as mentioned in the video description.
What the Video Doesn’t Tell You
However, it’s worth noting that there was no direct interaction between a person and the AI in the demo. Instead of Gemini reacting to real-time changes like predicting drawings or alterations on a table, the demo was created “using still images from the footage and prompting via text.” This approach in the video may mislead viewers about Gemini’s actual capabilities, especially since there are no disclaimers explaining how the inputs are generated.
Insights from Google
Oriol Vinyals, VP of Research & Deep Learning Lead at Google DeepMind, responded to the interest surrounding the “Hands-on with Gemini” video. In a post on X, he explained that Gemini was given sequences of different modalities—images and text—and responded by predicting what might come next. Vinyals encouraged developers to experiment similarly once access to Pro opens on 12/13. He clarified that the knitting demo featured in the video used Ultra.
Read More: Google’s Highly Anticipated AI Model, Gemini, Faces Delay Until 2024
Realism in the Video
Vinyals assured that all user prompts and outputs in the video were actual but shortened for brevity. He emphasized that the video aimed to illustrate what multimodal user experiences developed with Gemini could potentially look like, inspiring developers.
Building with Gemini
Vinyals suggested achieving comparable results for those building apps by instructing Gemini with a prompt that allows users to “configure” the model’s behaviour. An example is inputting “you are a science expert” before engaging in a back-and-forth dialogue. He shared a clip from AI Studio with Gemini Pro, expressing excitement about the progress since Flamingo & PALI and anticipating the innovative creations that people will develop with it.