If AI going from ‘cool tool’ to absolutely bonkers overnight blew your mind, you’re not alone.
In 2025, AI’s achievements — and yes, its hiccups too — felt like watching the last two years on fast-forward. Today, we’re giving you a rapid, no-fluff rundown of how far AI has come… and why this moment feels different.
First up: performance. In 2023, researchers rolled out brand-new benchmarks designed to push top AI models to their breaking point — think MMMU, GPQA, and SWE-bench. The idea was simple: “Let’s see where the ceiling actually is.” But two years later? That ceiling cracked. Scores jumped nearly 19 points on MMMU, almost 50 points on GPQA, and a wild 67 points on SWE-bench. AI went from “pretty good” to consistently impressive at hard reasoning, advanced coding, and multimodal tasks. And in some time-limited programming tasks, AI agents have straight-up outperformed humans. If you ask me, that's not gradual progress, that’s acceleration.
Second: AI is no longer living in the lab. It’s everywhere. In healthcare alone, the FDA approved 223 AI-enabled medical devices in 2023. By 2025, the total number of authorized AI/ML-enabled devices surpassed 1,250, up from 952 at the end of 2024. For context: that number was six in 2015. And on the streets, self-driving cars aren’t demos anymore — Waymo is clocking more than 250,000 paid rides weekly —and China’s Apollo Go robotaxis are right there with them, scaling fast and cheap.
Third: businesses are fully all-in. U.S. private AI investment hit $109 billion in 2024 — nearly 12× China and 24× the U.K. And in 2025, the numbers show continued, accelerated growth. AI adoption keeps climbing too. Large enterprises now lead with an 87% adoption rate, while mid-market firms have reached 75%. 2025 is also being called “the year of the agentic AI build-out”, and in many ways, it’s lived up to the name. Roughly 63% of enterprises increased their AI budgets this year, but here's the key part: research keeps showing productivity gains — not just for top experts, but for people earlier in their careers, which means, AI isn’t replacing expertise — it’s compressing the distance to it.
Now let’s talk geopolitics — because this part matters.
The U.S. still leads in top AI models by volume, but China is closing the quality gap fast. Benchmarks that once showed double-digit differences are now basically neck and neck. At the same time, AI development is going global — the Middle East, Latin America, Southeast Asia… you name it. This isn’t a two-player game anymore.
But here’s the tension. AI incidents are rising. Responsible AI standards exist… but adoption is uneven. Companies know the risks, yet fewer act on them.
Governments, meanwhile, are sprinting to catch up, rolling out regulations, frameworks, and massive investments, while tripping over themselves. In the U.S., it’s reached the point of open conflict between federal and state governments over who gets to regulate AI. Now, there’s already an Executive Order limiting state-level AI regulation, but let’s be real — that fight is far from over.
So yeah. We basically built the plane, launched it, and started arguing about seatbelts mid-flight.
Also... public sentiment is shifting too. In parts of Asia, optimism around AI is sky-high. In the U.S. and Europe? Still skeptical — but warming. Even historically cautious countries are increasingly moving toward AI. Now, AI anxiety hasn’t completely disappeared but curiosity is starting to win.
And here’s the quiet superpower story: AI is getting cheaper, smaller, and wildly more efficient. The cost to run GPT-3.5-level systems has dropped over 280-fold in just two years, hardware prices are dropping, energy efficiency is climbing and Open models are nearly catching up to closed ones.
In short, barriers are collapsing. But there’s a dark side: these systems guzzle electricity and water at alarming rates every single day.
The Big Takeaway:
AI isn’t just leveling up — it’s speed-running the whole tech tree. Every frontier is advancing at the same time, the distance between “best model” and “runner-up” is basically a rounding error, and the competition? Absolutely feral.
We didn’t crawl into the future; we teleported. A decade of progress got crunched into a handful of chaotic, glorious years. But while everyone’s celebrating benchmarks and bragging rights, the shadow side is getting louder too — jobs getting automated out of nowhere, cyberattacks scaling like a bad sequel, lone-wolf hackers doing in hours what used to take whole teams, and safety gaps that are still… very much gapping.
Going forward, fingers crossed we’re reporting more wins than wipeouts — but that’s on how these systems get built, guarded, and used.
So yeah: enjoy the holidays, recharge, touch some grass… and please, for the love of the internet, use AI responsibly.
