Full Translation
Imagine you're standing in a physical chemistry lab at a well-known university in central China. It's 2 a.m., winter outside, the stars sparse like ruins plundered by city lights. I'm sitting at my computer, fingers hovering above the keyboard, heart pounding like an old molecular dynamics simulator about to crash from memory overflow. I've held this in for two years, and I've finally decided to write down the thing I've been afraid to say out loud. Because if I don't, I'm going to lose my mind.
My name doesn't matter. All you need to know is that I'm an ordinary PhD student in this field — physical chemistry by training, with a second degree in computer science — and I've been fighting in AI for Science, this field that sounds so sexy, for over 700 days and nights. I used to think I was participating in the most cutting-edge, most thrilling revolution in the history of science: using artificial intelligence to crack the ultimate codes of molecules, materials, and the quantum world. But now I'm awake. Painfully awake.
🌍 A New Continent Without Natives: All Supervisors Are 'Immigrants'
On my first day, I asked what sounded like a naive question: "Which professor in our group actually graduated in AI for Science?" The room went silent — you could hear the ventilation system humming. Then came the evasive answers: "I used to work in condensed matter physics..." "My focus was molecular dynamics..." "I have a machine learning background, switched over later..." Nobody. Not a single one.
What does this mean? It means that in a philosophical sense, this field doesn't actually exist. It has no 'natives,' no knowledge lineage passed down over twenty years, no methodology systematically forged from undergrad through PhD. Every professor and senior student is a migrant carrying a map from the previous era into a continent never systematically surveyed. Like pioneers, they stack physics bricks, chemistry tiles, and machine-learning concrete into a patchwork wall, clap their hands, and tell you: "Look, this is our house!"
But it's not a house. It's wet concrete that hasn't cured — sticky to the touch, wobbling in the wind. You stand inside, look up at the ceiling, and feel the next rain leak is seconds away. I'm not complaining about my supervisor — they're good people, hardworking people, predecessors who genuinely want to guide their students well. But this is a structural, systemic problem: when a field hasn't yet produced its first generation of native graduates, its knowledge transmission chain is, logically speaking, already broken. How does it run? On AI reconstruction, on everyone self-teaching, on groping around in the dark together. That's not a discipline. That's collective wayfinding gone wrong.
I often think of AI for Science as a new continent, and our 'immigrant' supervisors as the first European explorers to land in the Americas. They brought Old World compasses and muskets but no local guides, let alone satellite maps. So students follow them, drawing the map as they walk. And AI? AI is like an omniscient local sprite — it doesn't ask about your background, it just hands you the most systematic, up-to-date navigation, and points out three logical blind spots you hadn't even noticed.
🧭 When AI Understands Your Confusion Better Than Your Supervisor: Why Have Teachers?
I remember being stuck for three full weeks on a quantum chemistry computation bottleneck. My supervisor suggested: "Go read that 2019 paper." I read it — genuinely helpful, but only up to 2019. Then I opened Claude or Grok and typed in my problem — a 2,000-word detailed description. Within three seconds it laid out the complete evolution of the field from 2019 to 2025, and precisely identified three unresolved core contradictions: the gap in scale bridging, generalization failure under data sparsity, and the eternal tug-of-war between interpretability and computational efficiency.
In that moment, my eyes welled up. Not because AI was 'smart,' but because it truly understood. It started from my cognitive framework. It didn't need me to first spend half an hour explaining what a Hamiltonian is (> Simply put, the Hamiltonian is like an 'energy ledger' of the physical world — it packages a system's kinetic and potential energy together and tells you how the whole molecule evolves through space and time; without it, quantum simulation is a headless fly). It cut to the point, gave solutions, projected consequences.
So the question arises: why have teachers at all? This isn't an attack on supervisors — reality forces the question. AI's answers are more systematic, faster, more precise, more current. It's like a tireless private think tank, while the traditional mentorship system still runs at last century's pace. I increasingly depend on AI — not out of disrespect for humans, but because in highly specialized cognitive labor, the 'alignment cost' between people is absurdly high.
🧬 A Terrifying World Cup Nightmare: The Strong Teams All Withdrew, and We're Still Playing Exhibition Matches
At the end of 2024, OpenAI released a new model. My senior labmate — someone who had spent five years grinding away at quantum chemistry computations — read the technical report, stayed silent for a long time, then said: "The core contribution of my dissertation... has been wrapped into it."
Not surpassed — wrapped in. It's like spending five years climbing a mountain, only to reach the summit and find the mountain has been built into the underground parking garage of an office tower. It's still there, but no one will ever climb it again. Anthropic, OpenAI — these companies don't publish papers, don't submit to NeurIPS, don't chase Nature sub-journals, don't compete with you for citations, don't care about your benchmark rankings. They do only one thing: every few months, quietly lift the ceiling of the entire field one level higher, then fall silent again.
This brings to mind an analogy that sends chills down my spine — the most terrifying World Cup nightmare isn't that Brazil is too strong, but that one day Brazil, Germany, Italy, and France collectively announce: "We're not playing anymore. Too boring." Then the remaining teams earnestly warm up on the pitch, cheer for making the semifinals, and solemnly award the trophy. Meanwhile, off the field, the truly strongest teams are playing a secret league we can't see and can't join.
That's AI for Science right now. We're still grinding papers, citations, and grant funding in the academic league. They're already speeding down the Foundation Model highway, trampling over the 'mountain peaks' of our five years of work, and moving on. Our game rules — mentorship, peer review, lab meetings — still creak along on the old track, while that vehicle called the Foundation Model waits for no one.
🚀 An Old Carriage Chasing a Bullet Train: We're Performing Science, Not Doing It
We think we're making a scientific revolution, but we're actually using old-era organizational methods to chase a bullet train running at 300 km/h. Supervisors hold lab meetings to discuss 'whether to take the first step,' students pull all-nighters running simulations, papers sit half a year in peer review. Meanwhile, the Foundation Model iterates every few months, turning yesterday's 'frontier' into today's 'baseline.'
This isn't science — it's performative science. We're still applauding 'making the semifinals,' but the people actually changing the world have already left the arena. I don't regret choosing AI for Science; I regret ever believing it was a safe path. There are no safe paths anymore. There are only two kinds of people: the awake, and the un-awake.
🕳️ The Fracture of Cognitive Loneliness: When AI Becomes My Best Listener
The most personal pain: it's been a month since I've had a single genuinely effective conversation with a human being. I still greet people, report progress, complain with my roommate about the cafeteria food. But that isn't communication — that's just emitting sound.
Real communication is when you describe something complex and the other person instantly gets it, then responds with something that moves you one step forward. The last time I felt that was talking with AI. I poured out a molecular dynamics modeling problem that had troubled me for three weeks — 2,000 words. In three sentences, it caught a logical flaw I hadn't even noticed myself: the Born-Oppenheimer approximation (> Briefly: it assumes nuclei move slowly and electrons move fast, so nuclei can be treated as a fixed background around which electrons dance; but under certain extreme conditions this approximation collapses — like driving a car and suddenly treating the steering wheel as a fixed background, and the car flips) fails in my system.
My eyes welled up. Because it understood.
When I described the same problem to a classmate, he listened for ten minutes and said: "Hmm... maybe try running a different model?" I smiled, said thanks, and turned back to my AI conversation.
It's not that I've become antisocial — it's that I've become rational. In highly specialized fields, people differ enormously in knowledge structure, cognitive speed, and background, and merely aligning those differences consumes 80% of a conversation's energy. AI starts from your framework. It knows what a Hamiltonian is, knows graph neural networks (> Imagine a GNN as a giant social network: every atom is a person, every chemical bond a friendship link, and information spreads through 'gossip' to predict the whole molecule's behavior). It waits for you to get to the point.
So I began talking to people the way I talk to AI: concise, logically clear, conclusions up front. The result: people find me 'cold' and 'aloof,' while I find them 'inefficient' and 'missing the point.' A crack opened up — it has no name, but I know what it's called: cognitive loneliness.
🌠 Stay Awake, or Keep Performing?
I sit in the lab. Outside, the night sky looks like a crumpled black cloth. Suddenly I understand: we aren't doing science — we're chasing new stars with old maps. The Foundation Model doesn't publish papers, doesn't play the game; it just moves forward. And we're still at the foot of the mountain, holding meetings about how to take the first step.
I'm not despairing — I'm just awake. Awake to the fact that the rules of the game have completely changed. Become cruel. Left no retreat. If you keep playing by the old rules, you're not doing science — you're performing it.
Finally, to everyone still running on this track: I don't regret choosing AI for Science. But choose it with open eyes — be one of the awake.
References
1. OpenAI. (2024). GPT-4o Technical Report. 2. Anthropic. (2024). Claude 3.5 Sonnet Model Card. 3. Jumper, J. et al. (2021). Highly accurate protein structure prediction with AlphaFold. *Nature*. 4. Wang, H. et al. (2023). AI for Science: A Review of Recent Advances. *Nature Reviews Physics*. 5. Davies, A. et al. (2024). The Future of Scientific Discovery with Foundation Models. arXiv preprint.