I'm making this prediction publicly and permanently because I want to be held accountable.
By the time this capsule unlocks, I believe we will have achieved (or be within months of achieving) artificial general intelligence. Not the narrow AI we have today, but systems that can genuinely reason, learn, and adapt across domains the way humans do.
Here's my reasoning:
THE SCALING HYPOTHESIS IS CORRECT
The evidence is now overwhelming. Each order of magnitude increase in compute and data has produced capabilities that weren't just incrementally better; they were qualitatively different. GPT-4 didn't just know more than GPT-3; it could reason in ways GPT-3 simply couldn't. We're seeing emergent capabilities that weren't programmed, weren't expected, and sometimes aren't even fully understood by the researchers who created these systems.
THE INFRASTRUCTURE IS IN PLACE
The major labs (OpenAI, Anthropic, Google DeepMind, and others) now have access to compute clusters that would have seemed like science fiction five years ago. The investment flowing into this space is unprecedented. We're talking about training runs that cost hundreds of millions of dollars becoming routine.
THE ARCHITECTURE BREAKTHROUGHS ARE HAPPENING
Transformers were just the beginning. The innovations in the past 18 months around sparse attention, mixture of experts, retrieval augmentation, and chain-of-thought reasoning are compounding. Each improvement enables the next.
WHAT I'M NOT SAYING
I'm not saying this will be safe. I'm not saying this will be beneficial. I'm not saying we're ready for it. I have deep concerns about alignment, about concentration of power, about the economic disruption that will follow.
But I believe it will happen. And when this unlocks, I want there to be a record that at least one researcher saw it coming and said so clearly, while it was still possible to prepare.
If I'm wrong, I'll have to live with that very public failure. But I don't think I'm wrong.