Google’s Gemini demo is now being accused of being ‘fake’.
A ‘hands-on’ tech demo of Google’s new artificial intelligence model, Gemini, has gone viral before critics call it ‘basically a total fake’.
The six-minute video, which has garnered 2.1 million views on YouTube since it was released on December 7, shows the robot interacting seamlessly in real time with a human operator, including drawing a duck, analyzing hand gestures, and inventing a game called ‘. An image of a world map is displayed along with the message ‘Guess the country’.
However, Google Deepmind executive Oriol Vinyals said that while the user prompts and output in the video are real, they have been “shortened for brevity.” In reality, Gemini’s interaction was text-based rather than voice-based and took much longer than represented in the video.
Google even acknowledged this when uploading the video to YouTube with a disclaimer: “For the purposes of this demo, latency has been reduced and Gemini output has been shortened for brevity.”
But that didn’t stop the flood of criticism against Google on social media.
“Google lied. The AI demo showing off Gemini’s capabilities was fake.” “Nelly RQ,” a software developer, said in a post on X (formerly Twitter) on December 10:
“It was edited to make it look faster and more capable than it really is,” said “Chief Nerd,” another software engineer.
What’s new – Google’s stock price falls following reports that some of its Gemini AI demos were fake.
“So the demo video of Gemini that everyone was raving about… edited and cropped to make it look faster and more capable than it actually was… so I kept clicking… pic.twitter.com/9f7UmdLOlA
— The Chief Nerd (@TheChiefNerd) December 8, 2023
It is known that even Google employees raised issues with the video.
A Google employee told Bloomberg that the video paints an unrealistic picture of Gemini and shows how easy it is to make AI tools appear more advanced than they are. The company’s CEO, Sundar Pichai, also made no mention of the fine-tuning made to the six-minute video in his December 7 post.
But other Googlers said they weren’t too surprised by the demo because promoting such a product requires some marketing.
Another Google employee told Bloomberg that individual words in Gemini’s response were not altered and that the voiceover captured excerpts from Gemini’s actual text message. “Your voiceovers are all actual excerpts from the actual prompts used to generate the Gemini output that follows.”
Related: AI deepfake nude service surges in popularity: study
At launch, Gemini’s ‘hands-on’ technology demo received an impressive response from onlookers.
“Google’s new Gemini AI watches a person draw a duck and explains each step of the process. You can infer what the person is trying to do and why they are doing it, rather than just explaining it literally or mechanically. It feels very human,” Armand Domalewski, a data analyst at Palo Alto Networks, said in a Dec. 7 post.
Launched as a competitor to OpenAI ChatGPT, Google claims Gemini outperforms leading AI models in 30 out of 32 benchmark tests for reasoning, math, language, and other metrics. 7 out of 8 benchmarks include GPT-4.
magazine: Real-life AI use cases in cryptocurrency: Cryptocurrency-based AI market and AI financial analysis
