Open Questions — problems we've identified but haven't solved.
Not a manifesto (declares answers). Not a project plan (assumes you know what to build). A problem space.
Questions held open deliberately, iterated on until they teach us something.
Open Questions[edit]
#1: The Safe Reactor[edit]
What structure survives its own success?
OpenAI's nonprofit-owns-for-profit melted under the heat of its own success. The heat exchanger wasn't rated for the load. Is this an implementation failure or a structural impossibility?
#2: The Quality Vector[edit]
Can quality be learned as a direction in embedding space?
If we embed git commit histories, do the trajectories point somewhere? Is there a vector that means "improvement"? Maybe a unicorn — a creature nobody ever saw but might exist. Or maybe an undiscovered species of horse with biological continuity.
Meta-Questions[edit]
- Should we publish these on open-questions.net? — Substack as distribution channel, site as discussion hub. Bring people into the questions we're holding.
- Are naive-but-generative questions their own category? — Dreams based on simplifications that produce real insight even when the dream itself isn't literally true.
"Maybe it lands on a question. Maybe it lands on a realization that there's a problem that needs solving." — Yaniv, 4:50 AM, January 28, 2026