Impressive demo of Gemini, the AI ​​from Google that wants to overthrow GPT-4

there The race for artificial intelligence The second passes. Google on Wednesday unveiled its most powerful AI model, called Gemini, that outperforms AI GPT-4 reference In many tests, the California-based company confirms. A lighter version is immediately integrated into its smart assistant cold And the Pixel 8 Pro smartphone – but not in the EU at the moment. However, we’ll have to wait until the beginning of 2025 to find out if the more powerful version of the Gemini Ultra might actually take over. OpenAI. Depending on what capabilities are revealed in the video, if Google doesn’t cheat during editing, it’s possible that the promises will be kept.

Gemini is a “large language model” (LLM) called “multimodal”. Translation: Able to understand text, images, and video clips and respond to them in writing or orally. According to Google, the super version has for the first time outperformed humans on the MMLU test, which measures comprehension and reasoning skills in multiple domains (language, mathematics, history, physics, medicine, law, and ethics).

Understanding and deduction

In a 5-minute demo, the user displays objects, drawings, and videos of Gemini. The AI ​​system verbally comments on what it “sees,” identifies objects, plays music, answers questions that require a certain degree of analysis, and justifies its “reasoning.”

He quickly recognizes a drawing of a duck, and later concludes that it is a plastic toy. Gemini then invents a game where you have to guess the country from a series of emojis. All you have to do is indicate the country and the device will recognize it. She also deals with the bonnet and then suggests things that can be knitted from two balls of wool in different colours. The AI ​​tries to guess the ending of a video in which a cat jumps on a shelf, thinking that the cat will succeed. It’s a failure, but it’s more the fault of the cat that crashed than the AI.

Google, like OpenAI, is working hard to achieve the holy grail: artificial general intelligence (AGI), which can reason as well as, or even better than, humans. Some experts speculate that it is simply a matter of power, but others are far from convinced, believing that machine intelligence is limited by its poor perception of the real world. Google intends to solve this problem by integrating its artificial intelligence into robots that take advantage of advanced sensory capabilities, especially touch.

BP with AFP