Social media users are asserting that the recently released video by Google showcasing the demo of their new model of AI (Artificial Intelligence) is misleading and embellished.
In the video, the new AI Model named “Gemini” is introduced to viewers. The AI was put into several tests such as spatial reasoning and logic, image sequences, magic tricks, cup shuffling, tool use, game creation, and coding to examine its capabilities which are undeniably impressive as another technological advancement. As of this writing, the viral video has over 2.3 million views and is generating mixed reactions from viewers.
While the majority were impressed, some were able to point out that the video was a form of dishonesty. Several viewers find it manipulative, with some saying that a full disclosure of how the video is generated is considered basic decency.
“We’ve been testing the capabilities of Gemini, our new multimodal AI model. We’ve been capturing footage to test it on a wide range of challenges, showing it a series of images, and asking it to reason about what it sees,” the video stated, claiming that they used a series of images and did not mention the use of text prompts.
Google has admitted that the video was edited. “For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity,” it states in its YouTube description.
Moreover, in a post that was published the same day that Google posted the video. In the blog, Google narrated specific details of how the demo was generated. Unlike what appears in the video where a demonstrator directly interacts with Gemini showing it real objects and moving pictures, the company admitted that they injected text prompts into the AI and only used still images.
In the video, it appears that the demonstrator shows a real rubber duck to the AI and squeezes it to make a squeaking noise where the AI was able to identify that it was a rubber duck. But in reality, the AI was only shown an image of a duck and inserted a prompt that creates a squeak when squeezed.
Another one is when the video shows how clever the AI was when the demonstrator played cup shuffling with the AI. It is a game where cups are placed in different positions hiding one ball underneath one of the cups where the AI was able to find the ball. However, the truth is, that the AI was given a text prompt about where the ball is.
More of the misleading clip is when the demonstrator shows a video playing a selected part of a film. Of course, the AI guessed the film and even specified what happens in the video. But again, it was far from the truth. The AI was shown a sequence of still frames and it was instructed to pay attention to the sequence.
Thus, the video and the blog show a contradiction. Parts of the video were even sped up just to make it look like the AI responded quickly. The video shows how AI answers to moving pictures and real objects but is actually shown a still image and fed with text prompts.
It is undeniably an impressive concept, but could do without the embellishment.
Other POP! stories that you might like:
Dog trainer Suzanne Clothier says AI was used to create a book in her name, without her consent
Model accuses fashion designer Michael Costello of whitewashing her face through AI
Snoozing on your alarm makes you mentally sharper, study claims
Still not over Asia Artist Awards 2023? Here’s a recap of its most unhinged and funniest moments