This page is still under construction.
| Theoretical Agent Foundations | Applied Agent Foundations | AI Alignment from Neuroscience | Improved Preference Optimization |
|---|---|---|---|
| Probability theory, decision theory, propositional logic, measure theory, theoretical computer science | Probability theory, formal logic, reinforcement learning theory | Software engineering, ML frameworks (PyTorch/TensorFlow), experiment design | Neuroscience fundamentals (neuroanatomy, fMRI analysis), computational modeling |
| Build and analyze idealized agents with provable value embedding; test theoretical guarantees. | Build and analyze idealized agents with provable value embedding; test theoretical guarantees. | Implement and scale agent architectures; validate in practical environments to measure robustness. | Translate brain-derived value signals into algorithmic objectives; test consistency with human behavior. |
| Formal MDP frameworks, reward-learning algorithms, safety proofs in toy domains. | Formal MDP frameworks, reward-learning algorithms, safety proofs in toy domains. | Scalable RL systems, large-scale simulators, distributed training pipelines. | fMRI decoding of value judgments, neural network models of decision-making. |
| Integrating theory of logic uncertainty or computational uncertainty with UDT setting, robustness under value uncertainty | • Extending proofs to high-dimensional state spaces• Bridging gap between theory and real-world noise | • Ensuring consistent reproducibility at scale• Balancing exploration vs. safety in dynamic settings | • Low signal-to-noise ratio in neurodata• Aligning neural correlates with computational objectives |