Gemini (language model)

2024

Gemini 2.0 Announcement

On December 11, 2024, Google introduced the next generation artificial intelligence model - Gemini 2.0. This universal neural network is capable of generating images, text and sound.

According to Google, Gemini 2.0 lays the foundation for so-called AI agents. These are bots that can perform certain actions on behalf of the user with high efficiency. An example is the Google Project Astra platform - a visual system that can recognize objects, help navigate the world and give clues about the location of forgotten objects.

𝓲

blog.google

In addition, Google has introduced Project Mariner, a new extension for the Chrome browser that can mimic user actions when working on the Internet. Another tool is the Jules AI agent, specifically designed to help developers find and fix low-quality software code. An agent based on Gemini 2.0 has also been created, which helps with the passage of video games.

Compared to previous generation AI models, Gemini 2.0 demonstrates higher performance and improved efficiency. If the Gemini 1.0 model was aimed at organizing and understanding information, then Gemini 2.0 is designed to make it much more useful. Google intends to use Gemini 2.0 in many of its products and services. In particular, the new neural network will be involved in the Google search engine, in the set of cloud services Workspace, etc.

Introducing Gemini 2.0, our most powerful model designed for the new era of AI agents. Thanks to advances in multimodality and the use of advanced tools, Gemini 2.0 is bringing the era of universal intelligent assistants closer, says Google Corporation.[1]

"Failed service" according to Sergei Brin

In early March 2024, Google co-founder Sergei Brin announced that the company "definitely messed up" with image generation in the large Gemini language model. Users complain about dubious answers and historical inaccuracies that the AI system allows in the process of work.

The Gemini bot has come under fire for generating a number of images said to have been deemed racist. Google suspended the corresponding function in Gemini, but after that users began to point out errors in regular text responses. According to Breen, the company does not quite understand why Gemini generates such incorrect results.

𝓲

Unsplash

We were definitely wrong with creating images [through AI]. I think it was mainly due to not being thoroughly tested enough. This upset many people for good reasons, "Breen said.

Sergey Brin also stressed that Google is far from the only company that is trying to solve the problem of inaccurate results issued by AI services. According to him, hallucinations, or false responses to user requests, as of the beginning of 2024 remain one of the difficulties in the development of AI services.

Gradually, we are getting artificial intelligence systems to hallucinate less and less, but I would definitely be happy with a breakthrough that would reduce the level of inaccuracies to almost zero. But you just don't have to count on breakthrough achievements, so I think we will continue to take gradual steps to reduce the number of errors over time, "said the Google co-founder.

It is also noted that Brin continues to participate in Google's work in order to strengthen the company's position in the hyper-competitive AI market. These technologies, in his opinion, have huge potential.^[2]

Notes