Google admits AI viral video was edited to look better

Google has confessed that a viral video demonstrating the capabilities of its artificial intelligence (AI) model, known as Gemini, was edited to enhance its appearance. The video, which has amassed 1. 6 million views on YouTube, depicts an impressive interaction between a human and the AI, with the AI seemingly responding in real time to spoken prompts and video. However, Google has now disclosed that the reality was somewhat different. The company accelerated the AI's responses for the demonstration, and the AI was not actually responding to voice or video inputs at all.

In a blog post published concurrently with the demo, Google detailed the actual process behind the video's creation. The AI was prompted using still image frames extracted from the video footage, and text prompts. This information was initially reported by Bloomberg Opinion, and subsequently confirmed by Google to the BBC. A representative from Google stated that the 'Hands on with Gemini' demo video presents genuine prompts and outputs from Gemini, and was created to exhibit the range of Gemini's capabilities and to inspire developers.

In the video, a human interlocutor poses a series of questions to Google's AI while displaying various objects on the screen. For instance, the person holds up a rubber duck and queries Gemini about its buoyancy. Initially, the AI is uncertain about the duck's composition, but after the person squeezes the duck, eliciting a squeaking sound, the AI correctly identifies the object. However, the actual process was quite different. The AI was shown a still image of the duck and asked to identify its material. It was then provided with a text prompt explaining that the duck produces a squeaking sound when squeezed, leading to the correct identification.