The goalposts keep moving. Here's why that's the wrong game to play — and here are the receipts.
Every time AI comes up, someone says it.
"I'll believe it when AI actually solves a real problem. Cure cancer. End poverty. Do something that matters."
And if they find out I teach vibe coding — using AI to write code without memorizing syntax — the follow-up is usually:
"That's not real coding. You're just making it easier to be lazy."
I get it. A lot of the public-facing AI ecosystem revolves around convenience: faster emails, faster prototypes, faster marketing copy. To a lot of people, AI still feels like a productivity layer sitting on top of the same world problems we already had.
But here's the thing: the miracle already happened. Several times. And the people waiting for it missed every one.
First, the Honest Acknowledgment
Before we get to the evidence — the skeptics aren't entirely wrong.
AI is overhyped in many contexts. There are startups wrapping thin layers around APIs and calling it a revolution. There are developers shipping unmaintainable systems held together by prompts and optimism. Critics are right to question durability and long-term value.
But there's a difference between "some AI products are shallow" and "AI has not produced meaningful breakthroughs." Those are not the same claim. The first is often true. The second is increasingly difficult to defend.
Problem 1: Protein Folding — A 50-Year Grand Challenge
For half a century, predicting a protein's 3D structure from its amino acid sequence was considered, by serious scientists, to be essentially intractable. The CASP competition had been tracking incremental progress since 1994, measuring advances in millimeters per decade.
In 2020, DeepMind's AlphaFold didn't win CASP. It lapped the field so completely that the organizers described it as a solution to the problem. By 2022, AlphaFold had predicted structures for over 200 million proteins — essentially every known protein on Earth — freely available to any researcher. The work was awarded the Nobel Prize in Chemistry in 2024, with the Nobel Committee specifically citing it as solving "a fifty-year-old problem." ¹ ²
Researchers are now designing drugs for neglected tropical diseases that were previously unfundable because the protein structures were unknown. Malaria, sleeping sickness, antibiotic resistance — all actively being targeted using AlphaFold outputs.
The standard hater response: "But it didn't cure disease."
Correct. It solved the structural biology bottleneck that prevented rational drug design at scale. That's like saying the telescope didn't discover Neptune — it just made discovery possible. Dismissing AlphaFold isn't skepticism. It's ignoring Nobel-level work.
Problem 2: Fusion Plasma — Controlling the Uncontrollable
Nuclear fusion requires confining plasma hotter than hundreds of millions of degrees inside magnetic fields. The plasma is chaotic — it tears, kinks, and disrupts in milliseconds. Traditional control theory cannot adapt fast enough. Human operators can't even come close.
In 2022, researchers at EPFL's Swiss Plasma Center and Google DeepMind trained a deep reinforcement learning agent to control the magnetic coils of the TCV tokamak reactor in real time. The AI learned to sustain stable plasma configurations, shape the plasma into forms physicists had only theorized, and adapt to real-time disturbances faster than any classical controller. Published in Nature (Degrave et al., Nature 602, 414–419, 2022) — not a press release. ³
The standard hater response: "Fusion is still 20 years away."
Maybe. But the plasma control problem — one of the hardest sub-problems in fusion — is now solved. No human or classical system could do this. AI did. That's not vibe. That's physics.
Problem 3: Weather Forecasting — 48 Extra Hours
For hurricanes, even a 24-hour prediction error means wrong evacuation orders, unnecessary economic damage, or death. Traditional models run on supercomputers for hours and still degrade sharply past a few days.
Google DeepMind's GraphCast, published in Science in 2023 and independently validated by the European Centre for Medium-Range Weather Forecasts (ECMWF) — widely considered the world's gold standard — outperformed traditional systems on 90% of over 1,380 tracked metrics, running in under a minute on a single machine. ⁴ For tropical cyclones, it consistently adds days of reliable lead time. In September 2023, GraphCast correctly predicted Hurricane Lee's landfall in Nova Scotia nine days in advance; traditional forecasts only locked in on Nova Scotia about six days ahead. ⁵
For climate-vulnerable nations, that margin is the difference between organized evacuation and chaos.
The standard hater response: "That's just pattern matching, not real physics."
The metric is lives saved, not philosophical purity. If a system outperforms physics-based models on real-world outcomes, the correct engineering response is to use it — not sneer at it.
Problem 4: Materials Science — 32 Million in 80 Hours
Finding a replacement for lithium in batteries would normally require testing millions of chemical combinations — a process estimated to take 20 years using traditional methods. Microsoft, working with the Pacific Northwest National Laboratory (PNNL), used AI to screen 32.6 million candidate materials in 80 hours, identifying 18 promising candidates and ultimately synthesizing a new solid-state electrolyte that uses approximately 70% less lithium than existing batteries. ⁶ ⁷
This wasn't solved by a human writing 32 million lines of test code. It was solved by defining parameters, constraints, and goals — and letting the model navigate the search space. The engineer provided the vision. The AI did the traversal.
Problem 5: Medical Imaging — Earlier Than the Experts
Multiple peer-reviewed clinical studies have shown AI systems matching or exceeding radiologists in detecting breast cancer, lung cancer, and skin cancer — not in theory, but in real hospitals with real patients. ⁸ A 2025 systematic review found AI demonstrated non-inferior or superior diagnostic accuracy compared to radiologists across breast and lung imaging, with additional benefits including reduced workload and improved triage efficiency. ⁹
Early detection is the single biggest factor in survival rates. AI isn't replacing doctors. It's giving them a second set of eyes that never gets tired.
The Moving Goalposts Problem
Here's what I've noticed about the "show me a real breakthrough" crowd.
Bring up AlphaFold: "That's just one protein database."
Bring up fusion control: "Fusion is still 20 years away."
Bring up weather forecasting: "Weather apps are still wrong all the time."
Bring up vibe coding: "That's not real coding."
There's no arrival condition. The bar moves every time AI clears it. That's not skepticism — skepticism has a falsifiable standard. This is something else: a prior commitment to dismissal dressed in the language of rigor.
Real skepticism sounds like: "Here's what would change my mind." And then actually updating when that thing happens.
Where Vibe Coding Actually Fits
The haters will say: the problems above were solved by PhDs using deep reinforcement learning and graph neural networks. That's real engineering. Vibe coding is people prompting their way to a CRUD app.
Fair distinction. I'm not claiming vibe coding and AlphaFold are the same thing.
But they're operating on the same principle: AI removing barriers that used to be considered permanent.
For AlphaFold, the barrier was computational complexity. For fusion control, it was real-time physics. For vibe coding, the barrier is access.
And access barriers are real barriers. For decades, learning to code required significant time investment before anything worked, tolerance for abstract concepts with no immediate payoff, access to teaching resources that varied wildly by geography and economics, and persistence through a culture that often treated confusion as a character flaw. Brilliant people with real problems to solve bounced off that barrier and walked away.
AI didn't make coding easier for experts. Experts were already in. AI opened the door for everyone else.
Today, someone with no background can describe what they want, receive working code, run it, modify it, and build something real. That is a solved access problem. It's not as photogenic as a protein structure database, but the mechanism is identical: a barrier that held for decades just came down.
The Internet Analogy
Nobody held the internet to a "solve a real problem" standard before deciding it mattered.
They looked at what it put in their hands — email, search, the ability to find information that previously required a library or a specialist — and decided that was enough. The miracle wasn't a single breakthrough. It was access, at scale, to things that used to require significant resources or expertise. Some of it was noise. Some was speculation. A lot of early websites were trivial. But underneath the chaos, foundational infrastructure was being built.
That's what today's AI moment looks like. Some of it will disappear. But underneath it, genuine advances are compounding quietly in science, medicine, engineering, and human productivity.
Society rarely recognizes transformation while it's still occurring.
The Bottom Line
Five impossible problems, already solved or in active progress:
- Protein folding — drug discovery accelerated by decades, Nobel Prize in Chemistry 2024
- Fusion plasma control — clean energy barrier cleared, published in Nature 2022
- Hurricane prediction — days of additional lead time, lives saved today
- Materials science — 70% lithium reduction identified in 80 hours vs. 20 years
- Medical imaging — cancer detection matching expert radiologists in real hospitals
If your bar for "AI is real" is a single magical cure for all of humanity's suffering, you'll be waiting forever. That's not skepticism. That's science fiction.
If your bar is solving problems that experts explicitly called impossible — AI is already there. You just weren't looking at the right journals.
The proteins are folded. The plasma is stable. The storm is tracked.
And every day, someone who thought programming wasn't for them builds their first working app.
The miracle isn't one big event. It's a thousand of those moments, compounding quietly, while everyone waits for something more cinematic.
Curious what vibe coding actually looks like in practice? The first lesson is free at gregthevibecoder.com — no experience required.
Sources
Nobel Prize Committee. "The Nobel Prize in Chemistry 2024." NobelPrize.org, October 9, 2024. https://www.nobelprize.org/prizes/chemistry/2024/press-release/
Jumper, J. et al. "Highly accurate protein structure prediction with AlphaFold." Nature 596, 583–589 (2021). https://www.nature.com/articles/d41586-024-03214-7
Degrave, J., Felici, F., Buchli, J. et al. "Magnetic control of tokamak plasmas through deep reinforcement learning." Nature 602, 414–419 (2022). https://www.nature.com/articles/s41586-021-04301-9
Lam, R. et al. "Learning skillful medium-range global weather forecasting." Science 382, 1416–1421 (2023). https://www.science.org/doi/10.1126/science.adi2336
Google DeepMind. "GraphCast: AI model for faster and more accurate global weather forecasting." DeepMind Blog, 2023. https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/
Microsoft Azure Quantum. "Unlocking a new era for scientific discovery with AI: How Microsoft's AI screened over 32 million candidates to find a better battery." January 9, 2024. https://azure.microsoft.com/en-us/blog/quantum/2024/01/09/unlocking-a-new-era-for-scientific-discovery-with-ai-how-microsofts-ai-screened-over-32-million-candidates-to-find-a-better-battery/
Moseman, A. "AI Expands the Search for New Battery Materials." IEEE Spectrum, 2024. https://spectrum.ieee.org/ai-battery-material
Patel, K. et al. "A Narrative Review of the Use of Artificial Intelligence in Breast, Lung, and Prostate Cancer." Life 13(10), 2011 (2023). https://pmc.ncbi.nlm.nih.gov/articles/PMC10608739/
Khalid, S.A. et al. "Comparative Performance of Artificial Intelligence and Radiologists in Detecting Lung Nodules and Breast Lesions on CT and MRI: A Systematic Review." Cureus (2025). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12671463/
Top comments (0)