Let’s summarize Musk’s newly unveiled xAI’s Grok-3 beta version.

────────────────────────

1. Musk and xAI’s Challenge: Unveiling Grok-3
   ────────────────────────
   • Elon Musk, CEO, personally unveiled the Grok-3 beta version through a live demonstration on social media platform X. It’s claimed to possess superior performance and innovative features compared to existing AI models.
   • Blind test results reportedly show it outperforming models from OpenAI’s ChatGPT, Google, and DeepMind.

────────────────────────
2. Grok-3’s Key Features and Differentiating Points
────────────────────────
① Beta and Mini Versions

• Following pre-training completed in January this year, it’s currently undergoing data training and is available in beta and mini versions.
• The mini version reportedly showed potential, sometimes even surpassing the beta version in performance.

② Deep Search Function

• A ‘Deep Search’ function, capable of creating comprehensive analytical reports by referencing hundreds of sources and reports, is slated for official release soon.
• This function rapidly gathers and summarizes diverse expert knowledge, significantly enhancing competitiveness against search engines and information services.

③ Enhanced Computing Power and Logical Reasoning

• Significant improvements in logical reasoning, computing power, and adaptability have been reported compared to previous models.
• Emphasis was placed on its ability to self-reduce “hallucinations,” highlighting its capacity to self-correct inaccurate data.

────────────────────────
3. xAI’s Technological Infrastructure and Innovation Strategy
────────────────────────
• Grok-3’s development utilized the Colossus supercomputer and 100,000 Nvidia H100 GPUs:

• GPU capacity was doubled in just 92 days, demonstrating rapid expansion of large-scale computing resources.
• The AI was trained for a total of 200 million hours, resulting in more refined and accurate data processing.

• The establishment of its own data center and reinforcement learning system, incorporating synthetic datasets and error correction functionalities, is also noteworthy.

────────────────────────
4. Benchmark and Blind Test Results
────────────────────────
• Grok-3 outperformed other AI models in various test areas, including science, coding, and mathematics.

• It notably achieved scores exceeding 1,400 points in the early version, a significant improvement compared to other models scoring around 1,300 points.

• These results suggest a significant impact on the AI market when the official Grok-3 model, API, and features like ‘Deep Search’ are officially released.

────────────────────────
5. Future Updates and Service Plans
────────────────────────
• The official Grok-3 version is expected to be released within weeks, along with the official API and services incorporating the Deep Search function.
• While the voice mode function is still in its early stages, Musk expressed the possibility of its release “next week.”
• Additionally, a premium service called ‘SuperGrok’ is planned, offering advanced features through a subscription model, promising broad applicability across various AI-utilizing fields.

────────────────────────
6. Summary and Outlook
────────────────────────
• xAI’s Grok-3 has quickly gained a competitive edge in AI model performance and is projected to significantly impact the overall AI market.
• Through the expansion of large-scale computing resources, the introduction of innovative features like Deep Search, and continuous updates, Grok-3 is poised to become a powerful tool for delivering diverse expert knowledge quickly and accurately, surpassing a simple chatbot.
• Musk’s confidence in surpassing “all global models in performance” suggests continued interest and updates are expected.

*Source URL:
https://n.news.naver.com/mnews/article/092/0002363623?sid=105&lfrom=twitter&spi_ref=m_news_x