Inspired or deceived? Google showed a fake video at the Gemini AI presentation

December 9, 17:59 Share:

Google exaggerated the capabilities of Gemini AI (Photo: Screenshot from Google video)

Google has admitted that it showed an edited video demonstrating new capabilities of its Gemini AI model.

Presenting Gemini, Google released a video that amazed many. It shows how the user communicates with artificial intelligence by voice in real time. The chatbot offers surprisingly accurate reasoning by commenting on what it sees using the camera.. For example, it recognizes that the user is drawing a duck even before the drawing is completed, solves logic problems, or reveals the meaning of the demonstrated tricks. At the same time, AI provides short and clear answers very quickly – exactly as if the user was communicating with another person.

https://twitter.com/Google/status/1732467423654105330

Google said it actually edited this video. The description of the YouTube video states that “ for the purposes of this demonstration, the latency has been reduced and the Gemini data has been condensed for brevity.”

In a comment to Bloomberg, a Google spokesperson noted that the voice in the demo read out Gemini's pre-generated hints, which the model produced by analyzing still images, not video, as shown in the demo video. In addition, the video also does not indicate that this is a demonstration of an improved version of Gemini ( probably Ultra, which hasn't been released yet).

Read also:
I'll tear everyone apart and won't even get tired. Google introduced AI that is superior to ChatGPT in everything, but there is a nuance

While the smooth voice conversation shown in the video is a far cry from Gemini's actual capabilities, Google denies that there was any wrongdoing on their part.. Gemini co-CEO Oriol Vinyals says the video was created to “ to inspire developers,” and all the chatbot’s tips are supposedly real.

As NV Techno wrote, this week Google presented a basic AI model with multimodal capabilities Gemini ( short for Generalized Multimodal Intelligence Network). It was announced that it will be capable of solving a wide range of tasks related to text, images, audio, video, 3D models and graphics. However, for now we are talking only about working with text. Google claims that their model has outperformed GPT-3.5. the previous version of the AI chatbot from OpenAI.

Read also: ChatGPT is safe for now. Apple won't release its own AI chatbot in 2024 – analyst Fear of a competitor? Apple and Google ignored ChatGPT in their top apps for 2023. Happy birthday, chat. 10 revolutionary changes that ChatGPT brought to the world in its first year of life