Earlier this year, Elon Musk’s XAI launched Grocke 3 as “the smartest AI on Earth”, 10 times better than its predecessor with computational performance. But it seems, their ambitions for Groke AI are beyond making it smart only than other chatbots. For its next step, Musk wants to use Groke AI to re -write the entire corpus of human knowledge – it appears that it is available online – because of course, the most data we have is waste.

In a series of posts shared on X, Musk revealed that he wanted to withdraw Grocke AI using a newly revised dataset. “We will use Groke 3.5 (perhaps we should make it 4 calls), in which there is advanced argument, to re -write the entire fund of human knowledge, to add missing information and remove errors,” Musk wrote. “Then came back to it. In any foundation model too much garbage is trained on any data.”

Musk, who founded the XAI in 2023, was to compete with Openai and other major players in the race for artificial intelligence. With the launch of Groke, he claimed that his reactions in existing models such as chatgip are “waking bias” and ideological diagonal and therefore, he wants to ensure that his chatbot is free from these limits. Musk has also asked X users to help again write this data.

He said, “Please answer this post with divisive facts for @grok training. Through it I mean things that are politically wrong, but still factually true,” he said.

Grocke 3, the large language model developed by XAI, was released in February and has already been deployed as a direct opponent for chat and China’s Deepsek. This X premium plus is available to customers and is also accessible through the official platforms of XAI including a dedicated app and website. During its launch, Musk said that the Grocke 3 AI model is trained on synthetic data and AI is designed with mechanisms to avoid hallucinations- a word used by some chatbots to describe incorrect or misleading information output by some chatbots.

And since its launch, Musk has also promoted the performance of Groke 3, claiming that it performs better than other publicly available models. Grocke 3’s training backbone is allegedly Xai’s Kolosus supercomputer, built in less than nine months and operated more than 100,000 hours of NVIDIA GPU processing. The model employs techniques such as reinforcement training and depends on a synthetic dataset to improve both logic and factual accuracy.

Now the next planned update for Grocke, aka Groke 4 or Grocke 3.5, which is expected to come by the end of 2025, will possibly include advanced arguments and hope to redefine how the AI ​​model is trained. And recent comments by Musk show that instead of relying on the huge but flawed internet dataset, he intends to build a new AI model, which he sees as a cleaner and factual accurate information on the alleged ideological narratives.