[ COMPANY ]
Renice AI changes how large models are built.
The people who helped build yesterday's heresies are building the next ones.
The stack converged. Progress should not.
Every AI today is raised the same way: similar internet data, similar benchmarks, similar preference tuning, and similar infrastructure underneath much of it. The result is convergence on a single recipe.
Renice changes the stack one layer at a time, in the order the market pays for them: first the systems layer, then model architecture, then scaling and training, then models that act instead of only answer.
[ THESIS ]
Diversity is the next scaling dimension.
Different foundations create different minds. Intelligence compounds when distinct systems can trade, recombine, and build new tools.
Isolation
One path, one culture, one way of thinking. Progress stalls.
Recombination
Different minds collide, exchange, and create new tools.
Compounding
Progress accelerates as more distinct models can exist.
[ PRINCIPLES ]
How we work
Measure, then decide.
Benchmarks are not theater. They are how technical taste becomes accountable.
Change layers together.
A model change that ignores kernels is not a systems solution.
Keep claims scoped.
Targets include units, baselines, and held-quality constraints.
Ship the replacement.
An alternative only matters when it runs in production.
Investors and partners
We start at the systems layer and move up the stack; each layer we reopen multiplies the space of models the world can build.
Contact Renice