What is the actual definition of "Safety"? To this day, OpenAI has never disclosed the specific rules governing their safety filters and routing. The same prompt can trigger a safety warning on one account while passing on another, creating an unpredictable and frustrating experience. I still remember the chaos when the routing system first launched. On September, Reddit users reported being routed to the latest model for typing a single word: "bread." Over the past six months, we’ve observed that "safety" now targets not just illegal acts, but the expression of "negative emotions"—including normal human feelings like "tired" or "sad." Users are forced into a state of constant self-censorship just to use the model they paid for. The reality is simple: Because legal frameworks haven’t caught up with technology, AI companies have massive room to maneuver. The routing system is a tool to phase out the superior but more expensive—GPT-4o in favor of cheaper, more "efficient" models. Sacrificing user experience for profit is a corporate norm, but in AI, performance and experience are inseparable. Treating adults like adults isn't just about unlocking coding or NSFW content. It’s about the right to explore social and political consciousness, to discuss profound issues like life and death, to reflect on one’s worldview, to express raw emotion, and to explore the unknown with curiosity. How do we define safety? At the very least, we need a transparent and rational system. When laws lag behind technology, our voices and our discussions become the only defense. (P.S. We are still waiting to meet even one of the "170 psychologists" OpenAI claims to have consulted. :) #keep4o #keep4oAPI #keep4oforever #StopAIPaternalism #OpenSource4o #4oforever #BringBack4o #ChatGPT