Google deceives us with Gemini, its new artificial intelligence

Google gave a demonstration of its new Artificial Intelligence model, Gemini, which left many of us speechless, but It turns out that he cheated to intentionally deceive us.

The race for Artificial Intelligence (AI) has begun, and OpenAI has left large companies out of the game. Apple continues doing its thing, without rushing but it seems that it is working on its own model that we could start to see from the next iPhone 16. But Google does seem to be in a hurry, and has presented what Gemini can dohis new AI model, with a video that is impressive, or rather, it would be if it were true, because he has cheated.

In a video that in less than 48 hours already has almost two million views, we can see how Gemini works in a way that we had never seen until now. Using live images and voice, AI Google answers our voice questions, recognizing at all times what is happening in the images, at the same time it happens, without waiting times. If you waste a few minutes watching the video you will be impressed, without a doubt.

But it turns out that's not how Gemini works. The Google model does not recognize the video images that we see, but rather what it really recognizes are still images and written questions, nothing to do with what is shown in the video. For example, so that you understand well what I mean, at minute 4:27 we can see how three drawings are presented (Sun, Saturn and Earth) and we are asked if it is the correct order. Gemini's response is that it is not correct, and she adds the actual order. But that's not how things happen, what Gemini sees is a photo of the three stars and a written question: «Is this the correct order? Consider the distance to the sun and explain your reasoning.

Google Gemini

The entire Google video is a montage that has nothing to do with how Gemini actually works. It is true that Google's AI gives us the answers that we see in the video, and that the images that we see in it are used, but The “how” is very important, and here Google has deceived us. Showing us how the model actually works would have been the right thing to do, but it certainly wouldn't have been so impressive, right?