What we can state right now is that there is a continuous race between the world’s giants as to who will lead the way in this field. There is no doubt that Chat GPT is currently dominant, but Google is certainly trying to give its answer. Google Gemini is exactly the new trump card of this company.

What is Google Gemini?

Gemini is a new and powerful AI model from Google that can understand not only text but also images, video and audio. As a multimodal model, Gemini is described as capable of performing complex tasks in mathematics, physics and other fields, as well as understanding and generating high-quality code in a variety of programming languages.

It is currently available through integrations with Google Bard and Google Pixel 8 and will gradually be integrated into other Google services. Google called it ” our biggest science and engineering project ever .”

Users can access Gemini starting this month, and a more advanced version will arrive early next year. Gemini is trained on Google’s Tensor Processing Units (TPU) and is faster and cheaper to run than Google’s previous PaLM , making the model far more efficient.

Google will also launch TPU v5p, a newer version of the TPU system, which is designed specifically for data centers that need to train and run large models.

Gemini comes in three variants

Gemini comes in three variants — Nano, Pro and Ultra, built to meet different user needs. The Nano is intended for quick tasks on the device, while the Pro is a versatile version that serves as a mid-level. The Ultra is the most powerful of the three versions and will be available next year as it passes safety checks.

You can “get a taste” of Gemini Nano on the Pixel 8 Pro, while you can try Gemini Pro’s advanced text capabilities for free within Google Bard. A recent comparison showed that the most capable version of Gemini outperformed the GPT-4 on multiple-choice exams, math and some other benchmarks.

Much of the analysis boils down to the fact that Google has access to an incomparably higher number of high-end chips and that its model outperforms GPT-4 on a performance measure related to computer calculations known as FLOPS .

Every day, in every way…

Currently, Gemini’s most basic models are text input and text output, but more powerful models like the Gemini Ultra can handle images, video, and audio. And ” it’s going to get even more general than that ,” says Hassabis, Google DeepMind CEO. “There’s still things like action and touch—more like robotic things.” “Over time,” he says, “Gemini will gain more senses, become more aware, and become more precise and grounded in the process. “These models simply understand the world around them better.” These models are still hallucinating, of course, and still have biases and other problems. But the more they know, says Hassabis, the better they will be.

For more news and interesting stories, visit our blog page or follow our Instagram profile.