Goodfire is an AI interpretability research lab building tools to understand, debug, and intentionally design neural networks by surfacing the internal features (via sparse autoencoders) that drive model behavior. - View it on GitHub