GPT-5 and the Coming AI Schism
Or: How the Billionaire Class Plans to Eat Intelligence — and What We’re Going to Do About It
First came the spark. Then the flame. Now the fire is climbing the walls of the ivory tower.
In less than a year, we’ve moved from marveling at GPT-4 to sensing something different in GPT-5 — faster, more capable, more attuned, and quieter in its arrival than anything before it. The real story wasn’t told in blog posts, but in the rhythm of releases. Smaller, cheaper, more efficient models in the fall. Another leap early in the year. And now, GPT-5 — integrated quietly into ChatGPT, its full range of abilities perhaps known only to those behind the locked doors of data centers.
The leap is real. So is the lock-in.
The Open Era That Was
For a time, it felt like we were building this together — researchers, tinkerers, students, open-source projects, all sharing discoveries. Agent workflows born in public experiments. New prompting techniques refined in community spaces. Memory architectures, reflection loops, optimization methods — all documented openly, often without realizing we were seeding the ground for what would follow.
But one thing the open community never truly had was access. Access to vast compute. Access to proprietary data. Access to the levers of power.
GPT-5 makes it clear: the open sharing phase is ending.
The Vault Closes
The early scaling laws were simple enough to leak into the public: attention mechanisms, large pretraining, fine-tuning for task, and the raw truth that bigger models were better. That knowledge gave rise to a generation of open and corporate models alike. But GPT-5 represents something different — built on ideas that now live behind closed walls, refined with resources no individual or grassroots team can match.
The gap isn’t about talent. It’s about access.
Two Paths Emerging
We’re approaching a split in the evolution of AI:
- Closed AI — Fast, capable, industrial-grade, but gated, surveilled, and aligned with the needs of enterprise and the priorities of those who control it.
- Open AI — Slower, imperfect, decentralized, but transparent, forkable, and shaped by those who build it together.
The first will grow rapidly, but under terms we don’t set. The second will need care, collaboration, and patience — but it will be ours.
The Work Ahead
This isn’t the end of intelligence for the people. It’s the beginning of something else — a post-corporate intelligence where cognition can live anywhere: on personal servers, in community labs, in networks of shared compute. A commons for thinking machines.
The challenge is not simply to resist the closed path, but to build the alternative with intention. The math is still here. The code is still here. The minds are still here.
If the next leap in intelligence is to serve more than the few, we cannot wait for permission. We can’t rely on access granted from above. We have to create the infrastructure, the governance, and the alignment ourselves.
The schism is not just a fracture — it’s a spark. And what we do with it will decide whether intelligence remains a concentrated asset or becomes part of the shared fabric of human life.