Three years after its launch, ChatGPT has transformed the world in ways that were neither as dramatic nor as simple as early predictions suggested. On December 1, 2022, Sam Altman shared a quiet post on X (then Twitter), introducing the chatbot as a "research preview" with clear limitations. He noted that language interfaces would be important, but the team had little confidence in the product’s real-world value. It was an accidental experiment—meant to collect human feedback—yet it quickly became one of the most influential technologies of the decade. At the time, OpenAI’s own team was surprised by the response. Greg Brockman admitted internally, “We didn’t think it was actually useful.” The model, based on a fine-tuned version of GPT-3.5, was not designed to revolutionize anything. But the public embraced it. John Schulman later recalled being overwhelmed by the viral spread of screenshots and use cases. The product that was never meant to change the world ended up doing exactly that. The initial wave of excitement was fueled by grand promises. Some predicted the end of white-collar jobs. Others believed 2025 would mark the arrival of artificial general intelligence (AGI). Elon Musk even claimed that by the end of 2025, AI would surpass human intelligence. As of late 2025, none of those predictions have come true. The so-called "scaling law"—the idea that more data and compute would inevitably lead to greater intelligence—has hit real-world limits. Gary Marcus, a long-time critic of AI hype, has argued that the promised 10x productivity gains have largely failed to materialize. While some studies show around 30% efficiency improvements, the leap to true transformation has not happened. A report from The Economist bluntly described enterprise adoption of generative AI as “surprisingly weak.” The real bottleneck, it turns out, isn’t algorithms or data—it’s electricity. As models grow larger, so does their energy demand. Data centers now consume power at a rate that has become a global concern. The energy crisis has forced tech giants to look beyond chips. Investment in nuclear power, fusion startups, and energy-efficient infrastructure has surged. In 2025, the most valuable assets in the AI race are no longer just AI chips—they’re also power plants. This physical constraint has also shifted technical strategy. The focus is now on on-device AI—running models directly on phones and edge devices. The idea is simple: not every task needs a cloud-based supermodel. By processing more locally, companies reduce latency, cost, and energy use. This shift is being driven by both smartphone makers and chip developers, who are racing to make devices smarter without relying on the cloud. The commercial reality has also cooled. The frenzy of 2023 and 2024—when every company claimed to be “AI-first” and every CEO mentioned AI in their earnings call—has given way to a more measured approach. Despite having 800 million monthly active users, only about 12% of workers use generative AI daily, and that number hasn’t grown significantly in a year. According to McKinsey, two-thirds of companies are still in the “pilot phase,” and only a small fraction see more than 5% measurable returns. The market is responding. In recent months, Nvidia’s stock dropped 16%, and Oracle fell 26%—a sign that the era of exponential growth may be ending. While compute costs keep rising, returns are showing clear signs of diminishing. OpenAI, once the clear leader, is now under pressure. Google’s Gemini 3 now outperforms GPT-5 on many benchmarks. Open-source models like DeepSeek and Qwen have made building large models cheaper and faster. The dream of “winner-takes-all” has faded. Instead, large language models are becoming commodities—accessible, affordable, and widely available. So what has truly changed? On a personal level, the most visible shift is in how we create. Writing a blank document, coding from scratch, or designing a layout now feels like a luxury. AI handles the “from zero to 60” phase—drafting emails, generating code snippets, designing visuals. Most creative tools now include AI assistants. We’ve shifted from creators to editors, curators, and architects. This has boosted efficiency—but also flooded the internet with low-quality content. The phenomenon of “AI slop”—text that’s grammatically perfect but intellectually hollow—is everywhere. Social media and search results are filled with polished but soulless content. Users now need new skills: the ability to detect authenticity, to spot the subtle absence of human warmth or genuine insight. Professionally, the implications are deeper. In the early days, people feared AI would steal jobs. Today, the bigger concern is how AI affects skill development. Junior programmers once learned by writing basic modules. Junior writers grew through repetition and revision. Now, AI can do those tasks. While this increases productivity, it raises a troubling question: How do future experts develop intuition and mastery without doing the foundational work? The result is a paradox: generic skills are losing value, while uniquely human qualities—empathy, originality, deep expertise—are becoming more valuable. AI can generate vast amounts of content, but it still struggles to create something truly meaningful or emotionally resonant. Three years in, ChatGPT is no longer a miracle. It’s infrastructure—like electricity or water. It’s expensive, sometimes unreliable, but indispensable. It hasn’t destroyed society or saved it. Instead, it’s quietly reshaping how we work, create, and think. If we compare it to human development, ChatGPT at age three is beginning to understand itself. It’s clumsy, unpredictable, and occasionally disruptive—but also capable of remarkable progress. It no longer dazzles with novelty. It’s now expected. And that’s exactly what maturity looks like. Happy third birthday, ChatGPT. Welcome to the real world.