Google is under investigation for the video it made to show off its new artificial intelligence model, Gemini.
Recently, just a few weeks before the end of the year, Google introduced Gemini, which they consider to be their biggest and most advanced AI model. They shared a video to the public and media outlets to demonstrate its capabilities.
Video Shows Chats and Recognition Skills, Raises Questions About Real-Time Functionality
In the six-minute video, you can hear conversations between a user and a chatbot powered by Gemini. The video also highlights Gemini’s skill in recognizing pictures and objects, and it can tell the difference between them. Some cool things Gemini can do include verbally describing drawings of a duck and distinguishing between a drawing of a real duck and a rubber duck, among other examples.
The description on YouTube from the company briefly mentions that, “For the purposes of this demo, latency has been reduced, and Gemini outputs have been shortened for brevity.” However, this disclaimer is not mentioned in the actual video.
After the launch, the company confirmed to Bloomberg that the demonstration was not done in real-time. Instead, it used static images and provided text prompts for Gemini to respond to, as highlighted earlier by The Information. The author pointed out that this approach was quite different from what Google seemed to be suggesting in the video – the idea that a person could have a smooth voice conversation with Gemini as it watched and responded in real-time to the world around it.
“The video is a visual representation of the potential interactions with Gemini, using actual multimodal prompts and outputs from testing. We’re excited to see what people come up with when access to Gemini Pro becomes available on December 13.”
Déjà Vu for Google
While it’s common for demos to be edited, the revelations about Gemini’s performance bring a sense of déjà vu for the search giant.
Earlier this year, Google received criticism from both the public and Wall Street for what its own employees referred to as a “rushed, botched” demonstration of its AI chatbots. This incident occurred during the same week that Microsoft had plans to showcase its Bing integration with ChatGPT.
Recently, The Information reported that Google abandoned plans for in-person events to launch Gemini and opted for a virtual launch instead.
Google is currently in intense competition with Microsoft-backed OpenAI’s GPT-4, which has been the most advanced and successful model so far. This week, Google released a white paper claiming that Gemini’s most powerful model, “Ultra,” outperformed GPT-4 across various benchmarks, although the improvements were incremental.
Conclusion
Google faces scrutiny over its Gemini AI model, with questions arising from a video demonstration. The company’s disclaimer about edited content adds to concerns, as revelations echo past criticisms of rushed AI demos. The competition with Microsoft-backed GPT-4 intensifies, as Google defends Gemini’s performance in a recently released white paper.
