ChatGPT may start to get scary. Google showed the most powerful AI model

Google announced its new artificial intelligence model last night, which – as announced – is the most advanced scientific and research project in the history of the company. Not only will it completely revolutionize the capabilities of the Google Bard chatbot, but it has already beaten ChataGPT in the latest version of GPT-4 in tests. Meet Gemini, whose possibilities are much wider than we could imagine.

Google’s new artificial intelligence model is surprising and a little scary

Google boasts that Gemini is not a continuation of any previously conducted project, but a completely new project created from scratch, prepared as part of extensive cooperation between many teams at Google. From the beginning it was designed as the so-called a multimodal model, i.e. capable of accepting, recognizing and processing various forms of information – text, image, video, sound or code.

The recently released model available in the Gemini 1.0 version is divided into three variants – Gemini Nano for working on mobile devices, Gemini Pro for scaling very diverse tasks and the most advanced and efficient Gemini Ultra for carrying out extremely complex tasks. The latter has already been subjected to a number of tests and outperforms not only existing artificial intelligence models, but even human experts (more on that in a moment).

Google has already shown off videos in which Gemini was subjected to a number of easier and more difficult tests. Especially those showing the ability to recognize and analyze various images, shapes or fragments of recordings make a huge impression.

AI recognizes handwritten sketches, decides what is the best choice in the situation described in the image, advises how to use the materials shown in front of the camera or instantly recognizes the final image in a connect-the-dots task. The entire video, lasting over 6 minutes, is packed with dozens of tasks that Gemini can handle similarly well, but often faster than humans:

Google Gemini is more than just a chatbot. He even beat the experts

Of course, Gemini’s capabilities are not based primarily on recognizing drawings. Google also showed research results in which Gemini competed with ChatemGPT in the latest version of GPT-4. In a test examining several key capabilities of such models – general knowledge, comprehension, math and coding – Gemini Ultra outperformed on seven out of eight tasks.

Google also boasts that Gemini Ultra is the first language model to beat (human) experts in MMLU (mass multi-task language understanding) text, which primarily checks the precision of understanding and analyzing a question before answering.

Gemini can also recognize and understand text, image and sound at the same time, thanks to which – as its creators explain – it can capture the nuances contained in the information provided to it and answer questions about complex issues. What’s more, it’s great at understanding the process of solving complex math problems, so it can show you how to solve a math or physics problem rather than just giving you the result.

We will soon be able to test the capabilities of Gemini Pro in the Google Bard chatbot, which has just been updated with a new model. The new product will be available at launch in over 170 countries around the world. The more advanced Gemini Ultra will not appear in Bard until early next year. For now, the model only works in English, but Google specialists are already working on adding additional languages. Gemini is also to be implemented in the Google search engine, then in the operating system of Pixel smartphones, and then in the company’s applications and services – including: on the Gboard keyboard or Chrome browser.