Google unveiled Gemini 1.0, an AI model, through a captivating demo video on December 7, 2023, showcasing remarkable object and image recognition capabilities.
The video, amassing over two million views on YouTube, initially sparked global excitement. However, recent revelations indicate that Google manipulated the video to amplify Gemini’s abilities.
Reports, notably from Bloomberg, cast doubt on the authenticity of the video, challenging the accuracy of Gemini’s touted efficiency. Contrary to the impressive responses displayed in the demo, Gemini’s actual performance falls short of the portrayed excellence. The video suggested Gemini outperforming CHATGPT in various tasks, triggering widespread discussions.
Gemini showcased its prowess by humorously responding to a blue squeaky toy duck and swiftly identifying hand gestures in a game of “rock-paper-scissors.” Impressively, it even unveiled the hidden location of a paper ball under rearranged cups. Unfortunately, the veil of Google’s deception has been lifted.
Parmy Olson at Bloomberg initially raised concerns about the video’s accuracy, prompting Google to confess to manipulating the footage. The exaggeration of Gemini’s multimodal abilities was acknowledged, with Google emphasizing the intention to offer a glimpse into the potential of future AI advancements.
Olson’s findings were substantiated by images from Google’s blog post, revealing that Gemini was guided by text prompts and still images. Google admitted to editing the video for a more dramatic effect, including instances where Gemini struggled with immediate recognition, requiring guided prompts for optimal performance.
While Gemini successfully guessed the order of drawings depicting the Sun, Saturn, and Earth, it was later revealed that it received a hint about considering the distance from the Sun. This crucial information was omitted from the demo video, raising concerns about the transparency of Google’s presentation.
The controversy surrounding Google’s deceptive practices leaves people questioning the company’s credibility. Many perceive this incident as a self-inflicted wound, tarnishing Google’s reputation for honesty in product presentations. Despite the lack of concrete evidence supporting a pattern of exaggeration in Google’s products, the company has inadvertently set a precedent for future multimodal AI development.
By: Noel Nortei