In subsequent entries, the billionaire encouraged the observers to send “controversial facts, politically incorrect, but true”, suggesting that they would be used to catch and correction of alleged distortions in history. At the same time, he promised that the improved groc one would receive advanced reasoning and would be able to analyze history before he re -acquired the edited material.
According to Musk, the problem is not only LLM hallucinations, but also “ideological distortions” of their creators. In this spirit, he mocked “Woke Chatgpt” and claimed that he only wanted objectivity, although the guidelines he published clearly reflect his own beliefs.
Read also at Business Insider
Critics point out that This is not the first time the billionaire reacts to the unflattering answers of his AI. A few weeks ago, Grok had to explain from entries about “white genocide” and correct statistics about right -wing violence that Musk did not like.
Check also: Samsung wants to combine a watch, a doctor and a patient in one ecosystem. Customers are to be healthier
Encyclopedia by Musk
Technically, the project means Creating a private, filtered encyclopedia that could become the only source of knowledge for the next iteration of Grok. Such a return loop – the model is trained on a text previously produced by yourself – it is a simple way to escalate errors or completely silencing content that is not compatible with the creator's line. Despite the high costs, Musk has both capital and Starlink infrastructure, which makes the experiment with “own truth” as real as possible to conduct.
History knows similar ambitions. Alexandrian librarians sorted out texts by canon, encyclopedists of enlightenment wanted to cut off from scholastics, and Soviet commissioners removed all biographies from the dictionaries after subsequent purges. Each of these projects, however, revealed the political interests of the era, not a universal truth.
This time the warning flows straight from the AI researchers environment. “This is straight from 1984 (the cult novel of George Orwell – ed.)” – commented the idea of Gary Marcus, a recognized critic of LLM technology, adding that prescribing history for one person is a new generation censorship. Rumman Chowdhury, a former head of ethics in Twitter, notes that The largest corporations are already “tanning” the models for their own needsbut Musk is the only one to speak about it in public, which should make everyone aware of how fragile the idea of ”neutral artificial intelligence” is.
The fastest moods of the current times – Reddit – reacts with a predominance of pure fear. In the opening topic, the user “Lividnegotiation2838” states that Musk is just meeting the darkest warnings of Geoffrey Hinton about AI training on falsehood and military use of technology. Another Internet user mocks that Grok will become “Hitler 4.20”, and another is joking about the “Grok 1984” version.
The fears dominate that Musk will offer a product whose value will be controlled by sovereignty over fiction. Users also score “pride and arrogance to think that all human knowledge can be prescribed.”
Among hundreds of comments it is difficult to find really favorable voices. Apparently, even those who do not like competitive AI models are terrified of the scale of interference in sources and Musk's idea.
Read also: KPMG report. Digital challenges that will define the competitiveness of companies by 2027.
Muska can be treated seriously
The warning signal may also be the fact that some of the tips of Internet users for Musk immediately turned into controversial information and memes, which shows how thin the border between “incorrectness” and an open disinformation is. If the grok begins to treat such content as final, the scale of damage will not be limited to incorrect answers – The possibility of verification will disappear, because the original texts will be overwritten.
The risk for future generations is double. First of all, the centralization of knowledge in the hands of one corporation means that the arbitrary decision on corrections will become invisible to end users. What is now a blog note now may be the only version of the facts in the school textbook online.
Secondly, models trained on the edited body will begin to produce further texts themselves, which to external researchers It will get harder to drain from original manipulations.
The idea of purifying Musk's data is not a salvation, but a written revision without the participation of the scientific process and without reviewing the academic community. Open access to sources, transparent edition logs and the ability to compare the version – these are minimal conditions, so that the modern AI billionaire would not turn into a digital minister of truth. If anyone convinces the narrative that only one man can separate the truth from falsehood, just look at Reddit and researchers' opinions to see almost unanimous skepticism.
As long as large language models will be trained on private, original copies of reality, users should maintain a reflexion of distrust. It is not a real innovation GROK 4.0 with a built -in history erasing functionbut developing open, versioned databases, in which each error can be seen in the hand. The billionaire's perspective cannot become the only narrative of our time, but on the other hand: who will prohibit the rich, right?
Author: Grzegorz Kubera, Business Insider Polska journalist
