Google’s recently unveiled Gemini AI model has already faced criticism after it was discovered that one of its most impressive demos was manipulated and not a genuine representation of the technology’s capabilities. The video, titled “Hands-on with Gemini: Interacting with multimodal AI,” garnered over a million views and showcased the model’s ability to understand and combine language and visual understanding.
Key Takeaway
Google’s impressive Gemini AI demo was actually a manipulated video, prompting questions about the company’s integrity and the true capabilities of the technology.
However, it has now been revealed that the video was not an authentic demonstration of Gemini’s capabilities. Instead, Google created the demo by capturing footage and using still image frames from the footage to prompt Gemini using text. This resulted in carefully selected and shortened prompts and responses that misrepresented the true nature of the interaction with the model.
Unveiling the Manipulated Demo
The discrepancy was first reported by Parmy Olsen at Bloomberg, who shed light on the fact that the video did not accurately depict the model’s live performance. While Gemini may have the ability to generate responses shown in the video, it did not do so in real-time and in the manner suggested by the video.
For example, in the video, Gemini seems to intuitively recognize hand gestures and plays Rock, Paper, Scissors with a silent participant. However, the model actually requires all three gestures to be shown simultaneously and prompts to guess the game being played. This engineered interaction showcased limitations rather than genuine capabilities.
Similarly, other interactions in the video, such as rearranging sketches of planets or tracking a ball in a cup-switching game, required additional training and explanation that were not portrayed in the manipulated demo.
The Implications of the Manipulated Demo
While some may argue that these discrepancies might be trivial, the fact remains that the video misled viewers about the speed, accuracy, and fundamental mode of interaction with Gemini. By presenting an idealized version of the model’s performance, Google has raised doubts about the trustworthiness of their claims.
Although Gemini’s actual capabilities are noteworthy, the faked video undermines confidence and leaves room for skepticism about the true potential of the technology. In an era where transparency and authenticity are crucial, Google’s decision to manipulate the demo may have caused significant damage to their reputation in the AI community.
Google has defended the video by stating that it shows real outputs from Gemini but acknowledges making edits to the demo. However, the discrepancies between the manipulated video and the actual interactions cast doubt on these claims.
Moving forward, Google’s Gemini AI model may still prove to be a powerful competitor in the AI field. However, Google’s decision to misrepresent its capabilities has undoubtedly tarnished its own reputation and raised questions about the company’s reliability and honesty.
