Google’s recent demonstration of its AI model, Gemini, has come under scrutiny as it appears the showcased capabilities might be more edited than initially portrayed. The video, which gained 1.6 million views on YouTube, presents an interactive exchange where the AI responds in real-time to voice prompts and video cues.
However, Google has acknowledged that the responses in the demo were sped up for presentation purposes and, more notably, revealed that the AI was not responding to voice or video interactions at all. The company clarified in a blog post that the video was created by using still image frames from the footage and prompting the AI via text.
While the video shows the AI correctly identifying objects and even participating in a game idea generation, the reality is that the prompts were not as spontaneous as they seemed. For instance, when asked if a rubber duck would float, the AI was initially shown an image of the duck and then fed a text prompt explaining its material and the resulting sound when squeezed.
Similarly, when the AI seemingly invented a game called “guess the country” based on a world map, it was, in fact, following specific instructions provided in a text prompt. The discrepancy between the video portrayal and the actual prompts used to generate responses has raised questions about the transparency of such AI demonstrations.
Google defended the video, stating it was created to showcase the diverse capabilities of Gemini and to inspire developers. However, the revelation of the behind-the-scenes prompts has sparked a discussion about the need for transparency and accuracy in presenting AI capabilities to the public.
