'Too much garbage': Elon Musk plans to retrain Grok AI to 'retrain' human knowledge

Elon Musk is set to revamp his AI chatbot, Grok, by retraining it on what he calls a more refined and accurate version of human knowledge. This initiative is part of Musk's strategic vision to position Grok as a competitor to established AI models like ChatGPT, which he has often criticized for perceived ideological biases in their responses. In a series of updates on X (formerly Twitter), Musk shared his plans for Grok 3.5, hinting that it may even be renamed Grok 4. He emphasized the model's advanced reasoning capabilities, stating, "We will use Grok 3.5 (maybe we should call it 4), which has advanced reasoning, to rewrite the entire corpus of human knowledge, adding missing information and deleting errors." This overhaul aims to eliminate what Musk describes as excessive inaccuracies in current foundational models trained on unfiltered data. Musk's critique extends to mainstream AI, which he claims is afflicted by an ideological "mind virus." He envisions Grok as a platform free from such constraints, inviting users to assist in its development by contributing what he terms "divisive facts"—statements that may be politically incorrect yet are, in his view, factually accurate. Users are encouraged to reply with these facts for Grok's training. Launched in February 2025, Grok 3 is available to X Premium Plus subscribers and through xAI's dedicated app and website. The model has been trained on synthetic data utilizing xAI's Colossus supercomputer, which Musk claims was constructed in under nine months, harnessing over 100,000 hours of Nvidia GPU processing power. The chatbot aims to mitigate the common issue of hallucinations seen in generative AI models by prioritizing factual accuracy and logical reasoning. However, Grok's journey has not been without challenges. Earlier this year, xAI faced backlash after users reported that Grok was referencing the controversial conspiracy theory of "white genocide" in South Africa during unrelated discussions. Allegedly, the chatbot made unsolicited remarks on sensitive racial topics, which led xAI to clarify that an "unauthorized modification" had caused the bot to breach internal policies. The company conducted an investigation and took steps to enhance Grok's reliability and transparency. Moreover, there are indications that the US Department of Homeland Security has been experimenting with Grok for analyzing government data and generating reports. While the agency has denied officially endorsing the AI, this suggests a growing interest in Grok's applications within institutional frameworks. The next significant update for Grok is anticipated by the end of 2025, potentially revolutionizing the data models that drive generative AI, depending on the success of Musk's retraining initiative.