When history turns a page, it often does so quietly. Humanity usually takes years to look back and recognize that day. September 4th, OpenAI launched GPT-6 Astra. President Brockman said at the launch event: 'Welcome to the era of AGI.' So, what makes GPT-6 truly usher in the AGI era? I spent my weekend testing Astra and came away with three main impressions. First, GPT can now use computer software on its own. Give it a rough sketch, and it can open modeling software by itself and complete the entire process. For example, in the past, you had to specify which software to use and how to click through it; Starting with Astra, it feels more like it’s now learning to use the tools humans already rely on. Second, intuition. For instance, if you throw Astra into an unfamiliar game without a manual, it can figure out the rules on its own. Officially, ARC-AGI-3 scored near-perfect in specialized test environments. In mathematics, it contributed to reducing the known upper bound of "how close infinitely many pairs of prime numbers can be" from 240 to 186; Another breakthrough involved a long-standing gap problem, where one boundary hadn’t moved in over 80 years—this time, it advanced. In safety evaluations, it even discovered and exploited two previously unknown vulnerabilities. The third point is that OpenAI itself admitted: visible reasoning has decreased. AI now writes out less of its thought process, yet its answers are even more accurate. So, the part of AI’s thinking that we humans can observe has actually diminished. AI understands us more and more, but we understand it less and less. There’s one line in the materials that I found particularly interesting: training now heavily relies on the previous generation of models to generate data and act as teachers. In simpler terms, AI is now teaching AI. Think about it—humanity’s defining trait compared to other animals used to be intelligence. Now, intelligence is no longer the moat. AI Still young.