About Metanthropic
Metanthropic is an AI safety and research laboratory dedicated to building verifiable, interpretable, and safe artificial intelligence systems.
We build verifiable systems
AI will have a vast impact on the world. We aim to build frontier AI systems that are not just powerful, but reliable and fully interpretable. We reject the "black box" paradigm. Our research focuses on developing architectures where the reasoning process is transparent and the outputs are bound by verifiable constraints.
Safety is physics
We treat AI safety not as a compliance checklist, but as a fundamental science akin to physics. We conduct empirical research to discover the "laws" that govern neural network behavior. By understanding the underlying dynamics of intelligence, we can build systems that are safe by design, rather than trying to align them after they are trained.
Interdisciplinary collaboration
Metanthropic is a collaborative team of physicists, mathematicians, deep learning engineers, and policy experts. We believe that solving the alignment problem requires perspectives from multiple domains. We bring together the rigor of the hard sciences with the agility of modern software engineering.
Our Values
Our values guide how we work together and the decisions we make. They are the operating system of our laboratory.
Act for the global good
We strive to make decisions that maximize positive outcomes for humanity in the long run. We are willing to be bold in our research directions if it means ensuring that AGI is a robustly positive force. We prioritize safety over speed, and humanity over profit.
Do the simple thing that works
We take an empirical approach to problems. We care about the size of our impact, not the complexity of our methods. If a simple linear probe explains a model behavior better than a complex neural network, we choose the simple explanation. Clarity is the precursor to safety.
Be helpful, honest, and harmless
We are a high-trust, low-ego organization. We communicate kindly and directly. We believe that the culture of the lab is reflected in the safety of the models we build. We foster an environment where it is safe to speak up about risks.