A live demonstration showcasing Google's latest AI model, Gemini, initially hailed as an impressive technical feat, has now come under intense scrutiny from critics who allege it was significantly misrepresented.
The six-minute video, garnering 2.1 million views on YouTube since its posting on December 7, exhibits seamless interaction between the AI model and a human operator, encompassing tasks like analyzing duck paintings, interpreting gestures, and even inventing a novel app called "Guess the Country." However, Google DeepMind's Oriol Vinyals clarified that while the video's prompts and output are genuine, they've been shortened for brevity. Contrary to the apparent real-time interaction depicted in the video, Gemini's interaction was text-based and notably more time-consuming.
Acknowledging the video's alterations, Google posted a disclaimer on YouTube, stating, "For the purposes of this demonstration, latency has been reduced and Gemini output has been shortened for the sake of brevity." Nonetheless, social media erupted with criticism directed at Google for what many perceive as a misrepresentation.
In a Dec. 10 post on X (formerly Twitter), software developer Nelly R Q accused Google of deception, asserting, "Google lied. The AI demo showing off Gemini’s capabilities was fake." Similar sentiments were echoed by Chief Nerd, another software engineer, who claimed the video was manipulated to present Gemini as more advanced than reality.
Allegedly, even Google employees voiced concerns about the video's accuracy. One employee revealed to Bloomberg that the video exaggerates Gemini's capabilities, illustrating how easily an AI tool can appear more sophisticated than its actual functionality. However, another employee defended the video as a necessary marketing strategy to promote such a product.
Addressing the controversy, a Google employee clarified that while individual words in Gemini's response remained unchanged, the voiceover was extracted from the actual text prompt used by Gemini. The launch of Gemini initially received enthusiastic responses for its human-like inference abilities, with some praising its seemingly intuitive understanding of human actions.
Positioned as a rival to OpenAI's ChatGPT, Google claims Gemini surpasses leading AI models in 30 out of 32 benchmarks across various domains, including language, mathematics, and reasoning, even outperforming GPT-4 on 7 out of 8 benchmarks.
