Elon Musk Teases ‘Scary Smart’ AI—Grok 3 Launching Soon

Speaking at the World Governments Summit in Dubai on February 13, he described the artificial intelligence (AI) model as “scary smart” and claimed it had outperformed all previous versions in internal testing.

One of the key differences in Grok 3’s development is its training method. Unlike other AI models that rely on real-world data, this version was trained using synthetic data and a self-correcting system.

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

Sidechains in Crypto Explained EASILY (Animated)

SUBSCRIBE

ON YOUTUBE

According to Musk, this approach allows the chatbot to recognize and remove incorrect information, which improves accuracy and logical consistency.

A major advantage of Grok 3 is its ability to pull real-time data from X, Musk’s social media platform. Instead of relying on traditional web browsing, it gathers information directly from X’s content.

The chatbot also includes an “Unhinged Mode”, which xAI describes as designed to produce responses that are "intended to be objectionable, inappropriate, and offensive".

Building Grok 3 required huge computing power. Experts estimate it used around 200 million GPU hours—far more than China’s DeepSeek V3, which only required 2.7 million. The model was trained on xAI’s Colossus supercomputer, running 100,000 Nvidia H100 GPUs.

Although Grok 3 is nearly finished, some final adjustments are still needed. Musk compared this stage to the last steps of a construction project—small details that improve the final product. He also suggested that Grok 3 might be available before OpenAI’s rumored GPT 4.5 and GPT 5 updates.

Recently, the AI-driven search engine Perplexity AI announced a million-dollar Super Bowl sweepstakes. How can users participate? Read the full story.