SAN FRANCISCO, Feb. 18 (Xinhua) -- U.S. billionaire Elon Musk's artificial intelligence company, xAI, officially released its latest flagship AI model, Grok 3, on Monday night.

Grok 3, claimed by Musk as the "smartest AI on Earth," can analyze images and respond to questions like OpenAI's GPT-4o and Google's Gemini, and serves as the core AI model powering various features on his social media platform X.

According to Musk, Grok 3 was trained at a massive data center in Memphis, Tennessee, which houses approximately 200,000 graphics processing units, and was developed with 10 times the computing power of its predecessor, Grok 2. It is a "maximally truth-seeking AI," said Musk during a live-streamed presentation.

xAI claims that Grok 3 outperforms OpenAI's GPT-4o across math and science problem-solving benchmarks, such as AIME and GPQA.

Additionally, xAI has introduced two specialized models, Grok 3 Reasoning and Grok 3 mini Reasoning, designed for complex problem-solving. These models function similarly to "reasoning" models like OpenAI's o3-mini and DeepSeek's R1.

To prevent distillation, a technique to extract knowledge from an existing model, Musk mentioned that certain aspects of Grok's reasoning process are intentionally obscured.

In addition, xAI introduced DeepSearch, a feature designed to perform comprehensive internet and X-based research to generate informative summaries. This functionality resembles OpenAI's deep research tool.

Grok will gain a "voice mode" as soon as about a week from now, Musk said. "When Grok 3 is mature and stable, which is probably within a few months, then we'll open-source Grok 2," he added.

Musk introduced Grok, a family of AI models developed by xAI, nearly two years ago, positioning it as an edgy and unfiltered model capable of addressing controversial topics that other AI systems might avoid. Enditem

Related

Go to Forum >>0 Comment(s)