Matt Shumer’s sweeping essay, ‘Something Big Is Happening,’ recently caused a stir on X. In less than 24 hours, it passed 20 million views; by today, it has received more than 80 million. The author uses a striking analogy to explain the changes happening in AI: think back to February 2020, when the pandemic was just beginning. The stock market was doing well. Children were in school. You went to restaurants, shook hands, and planned trips. If someone told you they were stockpiling toilet paper, you would have thought they had spent too much time online. Then, within three weeks, the whole world changed. Offices closed. Children stayed home. Life became something you could not have imagined a month earlier.
That really does describe the state of the AI world today. Think carefully about the series of events over the past few days and weeks, and you will find that, in just a few days, some of the most powerful AI systems have once again undergone dramatic changes.
The arrival of GLM-5 has allowed it to leap ahead in programming and close in on the frontier. Seedance 2.0 has made the video industry feel, once again, the profound upheaval that lies ahead. Soon afterward, Qwen3.5, an open-source model, entered the upper ranks occupied by closed-source models and topped the leaderboards again. OpenAI’s Codex5.3 and Anthropic’s Opus4.6 are competing for first place in programming. Google’s Deep Think achieved programming performance ranked eighth in the world on Codeforces, with only seven people ahead of it globally. And today, Gemini 3.1 Pro, with twice the accuracy, has decisively surpassed Gemini 3 Pro—the model placed on a pedestal just three months ago—setting new state-of-the-art results and pulling far ahead. Meanwhile, AI systems we considered quite powerful only a few months ago have already become old products, taken down and retired.
In just two weeks, constantly changing leaderboards and ever-rising numbers have revealed what lies behind the AI race: the rapid expansion of what machines can do. For AI, a few weeks are like decades or centuries; a few months ago already feels like another world.
Think back to that winter three years ago. Siri was still Siri, Xiaodu was still Xiaodu, and GPT-3.5 had just been released. Large language models were in their infancy, yet they were already astonishing. At the time, the model knew almost nothing. It could not access the internet. Its language abilities were mocked. It could not recognize objects or tell lies. People believed machines could never replace human intelligence and that repetitive labor would be the first thing to go. Three years later, in the spring of 2026, it is more capable than the vast majority of people. Its programming ability is reaching ever greater heights, while its abilities in mathematics, logic, and language are surging. It thinks what you cannot think and does what you cannot do. It has permeated nearly every frontier field—finance, technology, education, and medicine—and is beginning to challenge the peaks of human knowledge, reaching places we have not yet reached.
In mathematics, Google AI has released six doctoral-level papers in succession, sweeping through 18 major research problems. In finance, an empirical research paper that a Stanford professor completed with AI in less than an hour caused a sensation for its exceptionally high accuracy. When GPT-5.2 Pro independently proves a conjecture that has stood for 45 years; when Deep Think meets the gold-medal standard at the International Mathematical Olympiad and achieves a similar result at the ICPC; when AI can quickly complete tasks that take human experts almost five hours, with that figure doubling roughly every five to seven months—when AI fully enters the deep waters of human knowledge, do those of us who have grown numb to the daily leaderboard changes feel a slight chill down our spines?
A year ago, one iteration took a month to several months. Now it takes only days to weeks. Yet we have nothing that can limit the speed of its progress…
The torrent of iterative evolution never pauses for humanity at any particular moment. Even when we think its current pace is already dizzying, it continues to accelerate wildly.
AI is taking part in its own research and development, and the loop of recursive improvement will soon close. When releasing Codex5.3, OpenAI said: ‘GPT-5.3 Codex is our first model to play a key role in its own creation. The Codex team used its early versions to debug its own training, manage its own deployment, and diagnose test and evaluation results.’ AI is being used to build itself. When the vast majority of OpenAI’s code is entrusted to AI, and when the Claude team develops Coworker entirely through AI programming, AI iterates on the next generation of AI, day and night, without sleep or pause, again and again. Academia has also formally taken up the challenge of self-evolution. For the first time, ICLR 2025 included a workshop on ‘Scaling Self-improving Foundation Models,’ exploring how model performance might continually improve without human supervision.
Looking back at earlier industrial revolutions and the third scientific and technological revolution, energy and power replaced older tools and allowed people to work more efficiently. This AI revolution is fundamentally different: what it replaces is the process of labor itself! As that process disappears, execution loses its weight and becomes commoditized. When the physical and mental effort of typing code, deriving basic equations, or even building conventional models is compressed without limit, humanity’s position in the production chain undergoes an irreversible shift.
In responding to the torrent of our times, Shumer believes the greatest advantage now is to move first: to understand it first, use it first, and adapt to it first. We cannot go against the larger trend, so we must learn to live and work with it, gaining that early advantage. Many people reassure themselves—or say with some arrogance—that without AI, people could not accomplish many things. But the fact is that the world can never return to a time without AI. AI will not disappear. The ultimate measure of a person’s ability is no longer that ability alone, but a nonlinear combination of individual ability and the ability to use AI. The height that combination ultimately reaches is your height. On this point, Shumer says: put your pride aside. That law-firm partner did not think using AI was embarrassing because of their senior position. The people who will suffer most are those who refuse to participate: those who dismiss AI as a gimmick, those who think using it diminishes their professionalism, and those who assume their field is special and immune. No field is special.
I have always believed that learning how to collaborate with AI, and how to use this combined capability to accomplish goals almost beyond individual ability, is an important enough question to pursue. Since AI is already here, we should no longer confine our thinking to working in isolation. We should look at larger undertakings, at our curiosity about the world, and at our passion for doing something significant. We have never felt this before: everyone stands on equal footing before AI. We have never stood so close to the same starting line of creation. In this turbulent, uncertain environment of creation, with few barriers and many intersecting transitions, we can release our enthusiasm, curiosity, and creativity, and unleash tremendous energy. This is a redistribution of capability!
The future is here. Do your best to feel what that future means, and then begin immediately! Waiting is the greatest waste of the talent that each of you possesses.
