Google’s Gemini demo is now getting accused of being ‘fake’

A “hands-on” tech demo of Google’s new artificial intelligence model Gemini has gone from being the talk of the town to being accused by critics of being “basically entirely fake.”

The six-minute video, which garnered 2.1 million views on YouTube since its release on Dec. 7, shows it seamlessly interacting with a human operator in seemingly real-time, including analyzing a duck drawing, hand gestures, and inventing a game called “Guess the Country” with just an image prompt of the world map. 

However, Oriol Vinyals, a Google Deepmind executive has since clarified that while the user prompts and outputs in the video are real, it has been “shortened for brevity.” In reality, Gemini’s interactions were text-based, not voiced, and took much longer than how it was represented in the video.

Google even admitted when uploading the video on YouTube, with a disclaimer noting: “For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity.”

This, however, hasn’t stemmed a flood of criticism toward Google on social media.

“Google LIED. The AI demo flaunting Gemini’s capacities was a FAKE,” said one software developer “Nelly R Q” a software developer in an X (formerly Twitter) post on Dec. 10.

“It was edited, it was cut to look like it was faster and more capable than it actually is,” said “Chief Nerd,” another software engineer.

Even Google’s own employees have reportedly taken issue with the video.

One Google employee told Bloomberg that the video paints an unrealistic picture of Gemini and shows how easy it is to make the AI tool look more advanced than it actually is. The firm’s CEO, Sundar Pichai, also made no mention of the finetuning that went into the six-minute video in a Dec. 7 post.

However, another Google employee said they weren’t too surprised by the demo as some degree of marketing is required to promote such products.

Another Google employee told Bloomberg that individual words in Gemini’s responses were not changed, and the voiceover captured excerpts from actual text prompting of Gemini. “The user’s voiceover is all real excerpts from the actual prompts used to produce the Gemini output that follows.”

Related: AI deepfake nude services skyrocket in popularity: Research

At its release, the Gemini ‘hands-on’ tech demo garnered impressive reactions from onlookers.

“Google’s new Gemini AI watches a guy draw a duck and explains what he’s doing each step of the way—not just literally or mechanically, it can infer what the guy is intending to do, why he is doing what he does. It feels…very human,” said Armand Domalewski, Data Analyst at Palo Alto Networks in a Dec. 7 post.

Launched as a competitor to OpenAI’s ChatGPT, Google claims Gemini exceeds leading AI models in 30 out of 32 benchmarks testing for reasoning, math, language and other metrics — including GPT-4 in seven out of eight of those benchmarks.

Magazine: Real AI use cases in crypto: Crypto-based AI markets, and AI financial analysis