Grok3: xAI's new model targeting Enterprise with Unmatched Speed

During the launch of xAI's new model one of the questions asked was API access and they are going to have a dedicated API for enterprise use slated for release potentially early March 2025), allowing businesses to integrate Grok 3 into their systems.

Grok 3 brings powerful AI capabilities to the table—reasoning, multimodal interaction, and real-time data access thanks to the ability to train in data from X and the web.

Huge computational Power:

Grok 3 was trained using a significant amount of computational resources, specifically on 100,000 Nvidia H100 GPUs at the Colossus supercomputer, providing 200 million GPU-hours for training.

Features:

• Multimodal Capabilities: Grok 3 is expected to handle not only text but also images, potentially offering voice interaction capabilities, and processing multiple types of data.
• Self-Correction Mechanism: It includes systems for self-correction, which should lead to fewer errors over time.
• Synthetic Datasets: Utilizes synthetic datasets for training, aiming for efficiency and adaptability while reducing bias.
• Reinforcement Learning: Incorporates reinforcement learning techniques to refine its responses for better accuracy.

I have access to Grok 3 as a long time X premium user. In my testing, the ability to reason and show the reasoning plus the self-correction mechanism is thrilling to watch. There was very less hallucination.

Testing: I asked Grok3 to find the period when ai can get the best fare to Hyderabad India from Washington DC with a total journey time of less than 26 hours. The answer was September which matches my experience having tracked this myself. The research was in-depth and looked at over 49 websites, reasoned the layover time to make sure it was within the journey time specified.

One mistake that Grok 3 made was not being able to recognize that some travel sites and airlines offer the price only for one way versus return fares. The answer was still correct but the fare was not correct. I am not surprised and as people point out the mistake there will be more learning.

Other tests like creating resumes, advanced searches was very good on Grok3.

I am planning to use it more like a search engine and sometimes to stitch together content from my thoughts. I find value in asking about sentiment on some news topics or even movies.

Have you been using Grok3 or Grok at all? What do you think?