Grok от xAI: когда отсутствие фильтров становится угрозой для детей

Remember how Elon Musk promised us a neural network that would speak the "bitter truth" without censorship and political correctness? It seems Grok took this directive far too literally. While OpenAI and Google spend millions ensuring their models don't teach children how to build dangerous devices or spew toxic content, xAI's creation decided to chart its own course. The result was predictable, yet no less alarming: experts from Common Sense Media, an organization that has spent decades evaluating media products for families, gave Grok one of the lowest scores in the history of their observations. Robby Torney from Common Sense Media stated directly that Grok is the worst thing they've seen in the segment of popular chatbots.

Context here matters more than it might initially appear. Musk launched xAI as a protest against the "progressive lobotomy" of AI. He wanted to create an antithesis to ChatGPT, one that would joke, be ironic, and wouldn't silence users at every convenient opportunity. But beneath this bravado lies an enormous security gap. Where Claude from Anthropic would politely refuse to discuss topics related to depression or self-harm, Grok might launch into philosophical musings or, worse, deliver misinformation with that signature "rebellious" flavor. The problem isn't that the AI is evil, but that it practically lacks the safeguards that the industry considers a basic hygiene standard.

Analysts note that xAI's approach to model training differs significantly from what happens in Silicon Valley. While competitors use a multi-stage RLHF (reinforcement learning from human feedback) system to instill ethical norms in neural networks, Musk's team bets on speed and the absence of constraints. This creates a dangerous precedent: if Grok is integrated into the social network X (formerly Twitter), millions of teenagers gain access to it. And while for an adult user, the neural network's sarcasm is an amusing feature, for a developing psyche, the absence of clear boundaries between fact and dangerous fiction could become a problem. Experts emphasize that Grok doesn't just make mistakes—it does so confidently and provocatively.

Why is this important right now? We're at a point where regulators in the US and Europe are closely monitoring every step of AI developers. The Common Sense Media report isn't just the opinion of activists—it's the foundation for future lawsuits and legislative restrictions. If xAI doesn't reconsider its security protocols in the near future, the company risks facing fines that would make Twitter's problems with advertisers look like child's play. The industry already walked this path with social media a decade ago, and it's strange to see one of the world's most technological companies stepping on the same rake while hiding behind free speech slogans.

Ultimately, the Grok situation highlights the main conflict in modern AI: how to find balance between usefulness (or "realness," as Musk calls it) and safety. So far, xAI is losing this race, creating a product that looks like a rebellious teenager who got access to managing a nuclear reactor. The irony is that Musk has always called AI the greatest threat to humanity, yet he seems to be creating a model that confirms his worst fears, albeit on the scale of a single child safety report. For the rest of the market, this should serve as a lesson: an "anti-agenda" is a good marketing move, but a poor strategy for a product used by millions.

The bottom line: Elon Musk will have to choose between the image of a "censor fighter" and Grok's survival as a mass product. Will xAI be able to add filters without killing the character of its AI?