xAI, the AI company founded by Elon Musk, is releasing its new version of Grok 3 on Monday, February 17. Musk calls the new AI “scary smart,” and it brings major improvements in reasoning, power, and adaptability.
Grok 3’s development was helped by xAI’s Colossus supercomputer, built in just eight months. Powered by 100,000 Nvidia H100 GPUs, Colossus gave 200 million GPU-hours for training—ten times more than Grok 2. This has allowed Grok 3 to process large data sets faster and more accurately.
Apart from increased computing power, xAI changed its training approach. Grok 3 now uses synthetic datasets, self-correction, and reinforcement learning to improve performance.
- Synthetic Datasets: These are artificially created data, helping to train the AI more effectively and protect privacy.
- Self-Correction: Grok 3 can now check and fix its own mistakes, making its answers more accurate over time.
- Reinforcement Learning: The AI improves by receiving rewards or penalties based on its actions, making its decisions better.
These upgrades should reduce errors, known as “hallucinations,” and help the AI verify information against trusted sources. Grok 3 will also learn and adapt through continuous self-evaluation.
Human feedback is also used in Grok 3’s training. This helps it understand context better and refine responses based on what users need.
Early tests show Grok 3 outperforms rivals like ChatGPT and Google’s DeepMind in complex reasoning tasks.
Elon Musk hinted that Grok 3 could be a major step forward, saying, “this might be the last time that an AI is better than Grok.”
