Gemini. Google showed a highly advanced AI tool. However, there are controversies.

Gemini: a new AI tool from Google

Gemini is not only a continuation of previous projects, but a completely new result of the cooperation of various Google teams. It is a multimodal model, designed for advanced tasks and capable of processing such information as text, image, video, sound. It also includes generating high-quality codes in Java, Python, C and Go languages, and optimization for Tensor (TPU) v4 and v5e processors, brings a significant advantage in terms of performance and speed of operation.

"Artificial intelligence is a chance to help everyone, regardless of where they are in the world. It will bring innovation, economic progress and knowledge and education on an unprecedented scale" - argued Google CEO Sundar Pichai.

Google Gemini in many versions

The new model is available in three versions: Ultra, Pro and Nano. Gemini Ultra is the most efficient model designed for advanced tasks. Its performance is expected to exceed even the capabilities of ChatGPT 4.0. The Pro model was created for scaling a wide range of tasks, while Nano - is intended for use on mobile devices.

Google also announces that the benefits of Gemini will be broad and available to users of all services. In the coming months, Gemini will be integrated with key Google services, such as search engine, Google Ads, Chrome, Duet AI, Pixel smartphone operating system, Gboard keyboard.

Gemini 1.0 is now available in many services and on various platforms, and from December 6, the Gemini Pro model is used in the Bard service in English in over 170 countries and regions. The model currently only works in English, but Google plans to soon expand the availability of the model and make it available in more places and languages. On December 13, developers and business customers will also gain access to the Gemini Pro model through the Gemini API in Google AI Studio or Google Cloud Vertex AI.

Google Gemini: new competition for ChatGPT?

In comparative studies with ChatGPT in the GPT-4 version, Gemini Ultra achieved better results in key areas, which include general knowledge, understanding, mathematics, and coding. This model also turned out to be the first language model that surpassed human experts in multi-task language understanding.

The impressive capabilities of Gemini were presented in the promotional video above - the problem is that Google missed the reality in it. The company admits that some simplifications were made to energize the material. For example, playing "rock, paper, scissors" really doesn't look so smooth. Before preparing the video, Gemini was taught to recognize hand layout and game context, and this whole - let's be honest - tedious process, was described step by step on the Google for Developers blog.

"All user commands and results in the film are authentic, but they have been shortened for brevity. The film illustrates what multimodal user experiences built using Gemini might look like. We created it to inspire developers" - Oriol Vinyals explained about the controversy surrounding the film on X, vice president of the DeepMind team.

So Gemini is currently not as advanced and intelligent as Google would like to present it, which has not escaped the attention of those commenting on the film. Many of them feel cheated by the company.