We are approaching cyber-superintelligence.
But our systems aren't ready. We must secure our model weights and infrastructure against three new threats: sabotage, escape, and theft.
There is a way forward.
Secure Acceleration:
A Cyberdefense Strategy for Superintelligence
We are releasing the Harbor cookbook!
We designed the Harbor task format to make minimal assumptions about what a task should look like. So we put together a set of reference implementations for building evals and optimizing agents with Harbor.
v1 has 9 task recipes + 2
The Foundation series was my favorite book growing up, building AIs that can accurately model the future and how decisions affect it will be hugely impactful.
Mantic used Tinker to RL gpt-oss-120b on judgmental forecasting; the result outperformed frontier models on event predictions. Combined with @_Mantic_AI's forecasting architecture, task-specific training takes us to the cusp of automated superforecasting.
