Theory of AI
Understanding AI is the most urgent and difficult science problem of our time. We hope to build a theory of AI, connecting the first principles to the practice and to the interpretations of their capabilities.
Our lab is interested in identifying scientific principles of both artificial and biological intelligence (what is a principle?). We work across many fields of science: AI, physics, neuroscience.
On the AI side, we work on:
On the neuroscience side, I work on
Understanding AI is the most urgent and difficult science problem of our time. We hope to build a theory of AI, connecting the first principles to the practice and to the interpretations of their capabilities.
Despite the strong capability of neural models today, training and using them requires huge energy, and it is still important to identify better algorithms that makes these model more efficient and perform better. For us, understanding implies control. We hope to closely combine our understanding of both artificial and biological intelligence to improve AI systems.
How to make sense of the impressive capabilities of the large language models?
Without theory, experiments are blind. Despite the huge amount of experimental data we have accumulated for the biological brain, we are still far from understanding it well. In fact, we understand them so little that most theories today are still wild guesses. Can we propose computational models of the brain? Can we identify their similarity and differences from the artifical neural networks?
For full list of publications and preprints, see publications. Below are AI generated summaries of Liu Ziyin’s works.