Interpretability tooling for safe and reliable generative AI models. AI interpretability research company