Understanding what a model has learned
Goodfire studies representations inside trained models rather than relying only on their outputs. It has applied this approach to an epigenetic model built by Prima Mente, identifying a class of potential Alzheimer’s biomarkers.
Turning interpretability into development tools
The planned environment will help users locate the internal components responsible for a behavior and target them for training or intervention. The Series B also funds foundational research and new scientific collaborators.