I Can't Predict The Future Anymore

All I see is fog.

When I was 13, I was hit by a car. The man who hit me made the foolish mistake of driving on the same road I was skating in. It wasn’t too bad. My foot was flattened like a Wile E. Coyote cartoon, and I ended up limping around for a couple of months, but I survived. Internally, however, something far more interesting happened. I understood for the first time that I wasn’t invincible.

Obviously, if you asked me before the accident, I also would’ve said I wasn’t invincible - but the knowledge was abstract. It was a nebulous ghost of an idea, not a concrete understanding. Then I got bumped by a Vauxhall Corsa, and my own fragility was realised. It was like I could touch it. I didn’t just know with my mind anymore, I knew with my bones.

This week, I’ve had a similar experience. For the first time, I know in my bones that the world 10 years from now is going to be totally unrecognisable.

If you’ve been paying attention, you’ll have heard that OpenAI recently revealed a new model, o3. o3 continues every AI company’s disastrous model naming track record. If you’re thinking “Wait, didn’t we just get o1? Where’s o2?”, you’re not being dumb - o2 is just a giant mobile network in the UK, so the name was taken.

o1 is massively impressive, and has transformed the way I work. A way, which was already transformed by GPT4 and GPT3 before it. Now I’m able to give it a thousand lines of code, and it can make changes in an instant. It does things for me I didn’t even know were possible - and astonishingly, it’s already ancient technology. o3 makes it look like Patrick Star if he was kicked in the head by a horse.

---

I knew it’d be sudden, but a right angle is a bit ridiculous.

---

This is the ARC-AGI benchmark. It’s a reasoning test for AI Models. Now, it’s not as simple as saying that if something nails this test, we’ve achieved AGI, but it is fair to say it’s a bloody hard test. In fact, not long ago many people in the field believed it would stand for many more years - and we’ve basically just completed it.

Here’s the SWE (Software Engineering) benchmark.

---

---

12% to 71% in 9 months.

We also saw a 12x improvement on the hardest maths benchmark.

---

“These are extremely challenging… I think they will resist AI for several years at least” - Terence Tao

---

Suffice to say, exponential progress is going brrr.

What’s really crazy about all this is that 2 years ago the best model we had access to was an idiot. When ChatGPT first released, everyone used it to write insulting poems about their friends and left it at that. The idea that AI could build a new website from the ground up, do advanced mathematics, or generate lifelike images was absurd.

Yet, here we are. Two short years down the line and my life has morphed into a daily conversation with a robot that solves my problems for me. There are many domains that I will now never be better than an AI at. What comes next? No, really - I have no clue.

If you asked me in 2004 to guess what 2014 was going to be like, I feel like I would’ve gotten close. Well, maybe not me. I was probably picking my nose with a stick at the time; but a reasonable adult could’ve guessed. Most people knew the internet would change things. It was after the .com bubble, and the truly enduring internet businesses were starting to appear. No one used atlases anymore, they printed off directions from Google. We were already buying things on our computers off eBay. Email made handwritten letters and fax totally redundant. It didn’t take a mastermind to put together that basically all business would end up like this, and it wasn’t a crazy idea to think that we’d have smaller computers in our pockets.

Compare that to now, however, and the difference is staggering. What does 2034 look like when in 2024 we’re multiplying AI progress in mere months. Surely, most jobs will be a distant memory, but that might be one of the smaller changes. What does it look like when we make more progress in mathematics in 10 years than we did in the previous 100? What is a computer that’s 100,000x more powerful than our current hardware capable of? Will we expand off the Earth? Will we go inwardly into perfect virtual worlds? Will o6 pave over the planet, and bury us under a global patio? It’s a mystery.

I’m suffering a timeline myopia. Change used to be slow enough that you could extrapolate into the future with decent accuracy, but we’re at a point now where all I see is a thick fog. Every year is a step into something more and more alien to me. Whatever waits round the corner is going to be so drastically different to anything I’ve known, and I can’t help but feel like I’ll eventually look at my early life the way I currently look at cavemen rubbing sticks together. It’s pretty exciting. Equally terrifying.