AI how about this: Ilon Musk's new uncensored neural network unveiled

On the morning of February 18, Ilon Musk unveiled Grok 3 - a new generation neural network, which Musk himself called "the smartest AI on Earth". A distinctive feature of the novelty - high performance in special programs, training on the supercomputer Colossus with 200 thousand chips and almost complete absence of censorship. What are the possibilities of the new technology - in the material "Izvestia".

What is Grok from Musk

In the morning on Tuesday, February 18, American entrepreneur Ilon Musk presented the third generation of the AI model Grok. According to him, it is the smartest artificial intelligence on the planet. During the presentation, xAI, the company developing Grok, showed performance tests, according to which the new AI model is more efficient than all the latest models of the most powerful neural networks: Gemini-2 Pro, DeepSeek-V3, Claude 3.5 Sonnet and GPT-4o. Measurements were made in three categories: programming, math and scientific computing.

Of the key innovations, Grok 3 was trained on the Colossus supercomputer using 200k Nvidia H100 chips. This allowed the third generation of Musk's neural network to become ten times more efficient than Grok 2, which Izvestia reported on in August 2024. The chatbot also received a Big Brain button for writing code and solving math problems, and a Think button to activate reasoning mode. During the presentation, Grok 3 wrote code to simulate the movement of a spaceship, calculating the route to Mars and back to Earth.

Grok 3 also got DeepSearch, which the company describes as a new type of search engine with a demonstration of capabilities typically available to AI agents - autonomous systems that solve problems without human intervention.

It analyzes websites and social media posts, and frames the answer as a detailed summary. The only feature they promised to add in a week is voice mode. According to the developers, it is still unstable, so it will be launched closer to the end of the month.

How the AI industry landscape will change

According to Timofey Voronin, deputy director of the STI Competence Center based at Moscow State University, the release of Grok 3 will intensify competition in the global AI market, as well as competition specifically between U.S. and Chinese AI technologies. The expert admits that further tests of the technology may yield results that will diverge from the words of the scandal-plagued entrepreneur.

This development is moving the industry forward, said Vyacheslav Beresnev, CEO of the Neurolab Center for Artificial Intelligence Technologies.

- Considering how recently the xAI team started and the result it shows, there is a strong, reasonably ambitious player on the market, capable of giving the user a free, open and ethical model, at least now we see the broadcast of just such intentions, - said the specialist.

Grok 3 demonstrates impressive abilities in programming and game creation. This model can not only write code, but also generate entire game projects. This advantage will be the basis for the launch of a separate game direction of the company xAI, where the neural network will write code for the game, convinced IT expert, lecturer of the Department of Cybersport, Faculty of Game Industry and Cybersport at Synergy University Daniil Arzhakov.

Ilon Musk himself said during the presentation of the first version of Grok back in 2023 that he was creating a chatbot with a "rebellious spirit". he separately emphasized the rejection of most of the ethical principles that other neural networks use.

- For Russia, repeating the Grok 3 concept with its rejection of moral and ethical constraints is not an optimal choice. Instead of creating universal "supermodels" that require huge computing power and budgets, we should focus on training AI in highly specialized areas using data obtained with the help of domestic experts," suggested Daniil Arzhakov.

According to Vyacheslav Beresnev, departing from ethical rules may yield results, but at the same time it increases risks. It is possible to lose the trust of users, and this will lead to a slowdown in the development of technology in general and increased strictness of regulators. In other words, those neural networks and chatbots that emphasize ethics will remain popular in the first place.

- Russian AI solutions are highly self-censored. The first versions of Alice GPT were great, they handled a lot of tasks perfectly. And then censorship appeared abruptly," said Vladimir Zykov, head of the Association of Professional Users of Social Networks and Messengers.

As the expert additionally noted, now not only the AI race begins, but also the price race for advanced models, so the cost will play not the least role.

- It is hard to say for sure, but it is obvious that the price war for advanced versions has started seriously. GPT's cheapest subscription is $200, Grok-3's advanced model is $30, and DeepSeek and QWEN are all free. That's why DeepSeek is at the top of various countries in terms of downloads. After all, why pay if Chinese solutions will suffice for many simple tasks," summarized Vladimir Zykov.

During the presentation, Ilon Musk defined a new mission for Grok - to understand the nature of the universe. Now the neural network is available on the X platform, and its basic functionality can also be tested via a separate website and a mobile application for iOS devices.