Google's Impressive Gemini AI Demo Video Was Somewhat Fake, Tech Company Admits

Google unveiled Gemini AI as the next-generation multimodal AI tool that can process images, videos, text, and more. During the launch, Google claimed that it was smarter than GPT-4 and even outperformed human experts in a series of knowledge and problem solving tasks.

Google

What did Google fake during Gemini demo?

While all that sounds fascinating, it appears that Google faked some parts of the Gemini demo video. For instance, the video below shows Gemini responding to a user's voice and interact with a user's environment. In the video, Gemini is asked to guess what the user is sketching on a Post-It note. At this moment, it correctly answers duck.

Also read: Google Unveils Gemini AI, Claiming Superiority Over GPT-4 And Human Experts In Problem-Solving

Then, a rubber duck is put on a paper atlas and we see Gemini's ability to identify where the object has been placed. In the video, Gemini can identify objects, find where things may be hidden, and more. In essence, this was Google's way of demonstrating how Gemini processes different forms of information while performing logical and spatial reasoning.

How do we know it's fake?

It appears that it was fake. The model was apparently not prompted using audio and its responses were only text-based. The results were also not generated in real-time. A Google spokesperson told Bloomberg that the video was created "using still image frames from the footage, and prompting via text."

Google

The person speaking in the demo video was actually reading out text prompts that were given to the model. The robot voice given to Gemini was also reading out responses that were generated in text. Google essentially displayed the best bits to make it seem that Gemini is able to respond in real-time without any hassle.

Also read: Newly Launched 'Humane AI Pin' Powered By GPT-4 Will Replace Your Smartphone

"For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity," the video's YouTube description reads. Oriol Vinyals, VP of research and deep learning lead at Google DeepMind, who also helped lead the Gemini project, stated that the video demonstrates "what the multimodal user experiences built with Gemini could look like."

Really happy to see the interest around our ¡°Hands-on with Gemini¡± video. In our developer blog yesterday, we broke down how Gemini was used to create it. https://t.co/50gjMkaVc0

We gave Gemini sequences of different modalities ¡ª image and text in this case ¡ª and had it respond¡­ pic.twitter.com/Beba5M5dHP

¡ª Oriol Vinyals (@OriolVinyalsML) December 7, 2023

In many ways, then, the demo videos show what Gemini may be able to do as it learns and becomes smarter. What do you think about Google making it AI tool look smarter than it is? Let us know in the comments below. For more in the world of technology and science, keep reading Indiatimes.com and click here for our how-to guides.