Simplex is an AI safety research program at Astera Institute’s Neuro & AGI division that aims to build a science of intelligence by developing rigorous, physics-inspired theories of the latent internal structure of neural networks and using them to better understand and control model behavior.
Endorsements made here support Astera Neuro & AGI.
Simplex is an AI safety research program at Astera Institute’s Neuro & AGI division that aims to build a science of intelligence by developing rigorous, physics-inspired theories of the latent internal structure of neural networks and using them to better understand and control model behavior.
Endorsements made here support Astera Neuro & AGI.
Project Details
Updated 05/18/26 · By grantmaking.aiSimplex operates within Astera’s Neuro & AGI division as a small AI safety research team focused on interpretability and a rigorous science of intelligence. The group develops mathematical frameworks for the geometry of internal representations in large neural networks, aiming to recover belief structures in real language models, extend these ideas to more complex cognitive tasks, and build tools reliable enough to matter for safety. By grounding its work in the physics of information and testing theories on modern models, Simplex seeks principled methods for monitoring, understanding, and controlling advanced AI systems.
People
Updated 05/18/26 · By grantmaking.aiCo-Lead
Discussion
No comments yet. Be the first to share your thoughts.