When Google’s Gemini tech demo was unveiled last week, observers praised it, but the tech company acknowledges that portions of it were simplified for “brevity.”
Once the talk of the town, a “hands-on” tech demo of Google’s new artificial intelligence model, Gemini, is now being called “basically entirely fake” by critics.
Since its release on December 7, the six-minute video has received 2.1 million views on YouTube. It depicts the robot interacting with a human operator in a fluid and seemingly real-time manner, interpreting hand gestures, drawing a duck, and creating a game called “Guess the Country” using only an image prompt of a world map.
The video has been “shortened for brevity,” but Oriol Vinyals, a Google Deepmind executive, has since clarified that the user prompts and outputs are real. Gemini’s interactions took far longer than the video depicted and were actually text-based rather than voice-based.
With a disclaimer stating, “For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity,” Google even acknowledged this when posting the video to YouTube.
Nevertheless, this hasn’t stopped a barrage of criticism directed at Google on social media.
“Liked on Google. In an X (previously Twitter) post on December 10, a software developer named “Nelly R Q” claimed that the AI demo showcasing Gemini’s capabilities was FAKE.
According to “Chief Nerd,” another software engineer, “it was edited, it was cut to look like it was faster and more capable than it is.”
There have reportedly been complaints about the video even from Google staff members.
A Google employee claimed to Bloomberg that the video presents an inflated image of Gemini and demonstrates how simple it is to make the artificial intelligence tool appear more sophisticated than it truly is. In a post on December 7, the company’s CEO, Sundar Pichai, also failed to address the adjustments made to the six-minute video.
Another Google employee, however, stated that they weren’t overly shocked by the demonstration because these products need to be promoted in some way.
Another Google employee confirmed to Bloomberg that the voiceover included snippets of Gemini’s actual text prompting and that individual words in Gemini’s responses were left unaltered. “All of the user’s voiceover is taken directly from the prompts that were used to create the resulting Gemini output.”
When the Gemini “hands-on” tech demo was released, observers responded admirably.
“Google’s new Gemini AI observes a person drawing a duck and explains each step of the process—not just literally or mechanically, but also able to deduce the person’s intention and motivation. Palo Alto Networks data analyst Armand Domalewski wrote on December 7 that “it feels…very human.”
Google launched Gemini to compete with OpenAI’s ChatGPT. According to Google, Gemini outperforms top AI models in 30 out of 32 benchmarks that test reasoning, math, language, and other metrics. In seven out of the eight benchmarks, GPT-4 is tested.