Elon Musk Says His New AI Chatbot Is &#039, Scary Smart&#039, —And Arriving in Weeks

Elon Musk announced that the upcoming technology of his company’s AI bot Grok may be weeks apart from being released, calling it “scary intelligent” and asserting that it had already outperformed all other AI models in screening.

The xAI CEO made these notes on February 13 at the World Governments Summit in Dubai.

” At days, I think Grok-3 is kind of terrible smart”, Musk said. ” It comes up with solutions that you wouldn’t even anticipate—you know, not obvious options”.

Grok-3 was trained specifically by the robot developers. Instead of using real-world information like ChatGPT, Grok-3 relied on synthetic data and employed a self-correcting method to maintain reasonable regularity. It got so correct, Musk claimed, that even when it encountered wrong information, the program reflected on the data and removed material that didn’t fit reality.

The mathematical demands for coaching Grok-3 were enormous. Authorities calculate that it required 200 million GPU days, dwarfing its Foreign company DeepSeek-V3’s 2.7 million hrs. It ran on xAI’s Colossus supercluster with 100, 000 Nvidia H100 GPUs—ten times more processing power than its predecessor. Yet without fine-tuning, Musk claimed the basic design performed better than Grok-2.

Grok-3 had the advantage of being able to peel the social media app in real time rather than use the web while Grok-3 was integrated with X, Musk’s social media platform. The program can take real-time data from X, and features what the business called” Unhinged Mode”— which, according to xAI’s personal Question, is “intended to be disagreeable, improper, and offensive”.

The method isn’t quite ready for prime time, nevertheless. The next 5 % of the house’s plaster, painting, and trimming are all done by Musky, who compared the remaining work to finishing it:” Even though it’s not much labor, it transforms the house.”

But, it may be released sooner than OpenAI’s GPT-4.5, at least, which Sam Altman said could be released in weeks or months.

” Probably ( Grok-3 ) gets released in about a week or two”, Elon said. He didn’t understand whether the fresh version may be publicly accessible or put behind a registration, as happened with Grok-2 at first.

Competition in the AI space has intensified. While ChatGPT dominated the market share in 2024, Chinese open-source model DeepSeek-V3 emerged as a serious contender, outperforming both GPT-4o and Meta’s Llama 3.1 despite using far fewer resources.

Grok was first made available on X Premium, which substantially limited its availability. After that, all users of Musk’s social media platform received a free copy of the website, which is now accessible to everyone else.

xAI enters reasoning AI battle

Major AI players are shifting their attention to reasoning models, creating AI models that can reflect on specific issues and discover solutions after a protracted and complex chain of thought reasoning.

The idea was first explored by Matt Schumer, back when Reflection 70b was announced. The model was taught to use Chain of Thought reasoning, and despite only being a Llama 70b finetune, it was supposed to outperform Claude 3.5 Sonnet in challenging situations.

That didn’t work, but just a few weeks later, OpenAI announced its” OpenA I o1″ reasoning model, applying that same concept effectively. As OpenAI’s moat, that model was seen as the new standard for the logical capabilities that AI models can exhibit, and it established a new standard for the field.

However, the release of DeepSeek changed everything. A group of Chinese researchers created a model that was superior to o1 for a fraction of the price, and it was also made open source.

Since then, OpenAI has announced that all of its future models will be combined into a single, top-notch AI that rejects the conventional GPT architecture and places a premium on deep reasoning.

xAI appears to be following the markets.

” Grok-3 has very powerful reasoning capabilities”, Elon Musk said.

He didn’t disclose additional information about the model’s structure. The current version of Grok-2 is placed in the 18th position in the LLM Arena, well below competitors like GPT, Claude, Gemini, Qwen or DeepSeek.

For future models with” trillions of parameters,” xAI intends to expand its computing infrastructure to 1 million GPUs. Musk believes that the ultimate objective is to develop more sophisticated models for general intelligence.