Google is facing scrutiny over its demonstration video for its newly launched AI model, Gemini.
On Wednesday, just weeks before the year’s end, Google launched what it considers its largest and most capable artificial intelligence model Gemini and presented a demonstrative video to media outlets and the public.
The six-minute video includes spoken conversations between the user and a Gemini-powered chatbot, and also shows Gemini’s ability to recognize visual pictures and physical objects and know the difference. Some of the capabilities were impressive, such as Gemini’s ability to voice aloud a description of drawings of a duck, and describing a drawing of a duck versus a rubber duck, among other examples.
The company’s description on YouTube includes a short line “For the purposes of this demo, latency has been reduced, and Gemini outputs have been shortened for brevity.” However, it doesn’t make that disclaimer in the video itself.
Following the launch, the company later confirmed to Bloomberg the demo wasn’t conducted in real time, but instead used still images and fed text prompts that Gemini responded to, as previously pointed out by The Information. The author noted that was “quite different” from what Google seemed to to be suggesting: “that a person could have a smooth voice conversation with Gemini as it watched and responded in real-time to the world around it.”
After multiple requests for comment, the company on Friday told CNBC, “The video is an illustrative depiction of the possibilities of interacting with Gemini, based on real multimodal prompts and outputs from testing,” a spokesperson said in a statement. “We look forward to seeing what people create when access to Gemini Pro opens on December 13.”
Though demos are often edited, the subsequent findings of Gemini bring up déjà vu for the search giant.
Google faced criticism from the public and Wall Street earlier in the year for what its own employees called a “rushed, botched” demonstration of its AI chatbots, which happened the same week Microsoft planned on showcasing its Bing integration with ChatGPT.
Earlier this month, The Information reported that Google scrapped plans for a set of in-person events to launch Gemini, eventually settling on a virtual launch.
Google is in fierce with Microsoft-backed OpenAI’s GPT-4, which has been the most advanced and successful model up until this point. Google this week released a white paper that claimed Gemini’s most powerful model “Ultra” outperformed GPT-4 against several benchmarks, albeit incrementally.
