Google’s Edited Demo Video for Gemini AI Draws Backlash

Not long after Google took the limelight with an impressive video demonstrating the capabilities of Gemini AI, the tech giant is facing criticism for using an edited video.

On December 6, Wednesday, Google rolled out Gemini AI - the company's largest and most capable artificial intelligence model. Following the launch, Google released a six-minute-long video demonstrating Gemini's capabilities. As it turns out, the actual process of the demonstration was quite different from what the video appears to suggest.

Google Admits That the Demonstration Wasn't Carried Out in Real-Time

The video in question demonstrated spoken interactions between the user and a chatbot powered by Google's new AI model. Demonstrating Gemini's capabilities in precisely recognizing both physical images and visual images, it showed the AI model audibly reacting to the sketch of a duck being drawn all the way from a squiggle to a complete drawing.

Gemini AI has outperformed various AI models, including ChatGPT 3.5, and is set to be integrated into Google Bard.

The user then proceeded to color the duck blue, following which the chatbot pointed out that it was an unusual color for a duck. The AI model then expressed surprise upon being shown a blue toy duck before stating that it was likely a toy.

The demo further went on to show off other capabilities of the AI model, such as recognizing shadow puppet gestures, tracking the cup with the ball in a cup-swapping game, etc.

However, the video's description included a short statement that said, For the purposes of this demo, latency has been reduced, and Gemini outputs have been shortened for brevity".

The fact that the disclaimer wasn't included in the video itself brought it under scrutiny and had already begun to draw criticism before other discrepancies were pointed out.

Just a day after the video was released, The Information pointed out that the researchers didn't actually prompt the AI model with regular speech and hand gestures as shown in the demo.

Instead, they fed Gemini text prompts and still images to generate the results. Gemini later confirmed that this was indeed the case and that the video hadn't been made in real time.

An Illustrative Depiction of the Possibilities", Google Claims

Following several requests for a comment, Google finally released a statement describing the video as an illustrative depiction of the possibilities of interacting with Gemini". It was based on actual multimodal prompts and test results, the tech giant assured.

As critics have pointed out, the actual capacities of the AI model don't seem to be nearly as impressive as the demo video made them out to be.

This isn't the first time that Google has faced flak for faking such demonstration videos. In the past, questions arose around the legitimacy of Google's Duplex Demo - a demonstration video showing its AI assistant Duplex being used to make restaurant and hair salon reservations.

Following the demonstration, many experts and journalists concluded that the demo was actually a set-up. Several media reports accused Google of faking the calls and tasks executed during the demonstration based on suspicions arising from factors like background noise during the phone calls.

The post Google's Edited Demo Video for Gemini AI Draws Backlash appeared first on The Tech Report.