Generally Intelligent

There will be a scientific theory of deep learning

Malleable software and human agency with Geoffrey Litt

Matt Boulos, Imbue: Technology as power

Episode 37: Rylan Schaeffer, Stanford: On investigating emergent abilities and challenging dominant research ideas

Episode 36: Ari Morcos, DataologyAI: On leveraging data to democratize model training

Percy Liang, Stanford: The paradigm shift and societal effects of foundation models

Foundation models mark a fundamental shift in AI, necessitating rigorous, standardized evaluation frameworks to ensure these systems remain tethered to reality rather than becoming engines of polarization. As these models increasingly mediate human interaction with information, independent academic research serves as a...

Episode 34: Seth Lazar, Australian National University: On legitimate power, moral nuance, and the political philosophy of AI

Episode 33: Tri Dao, Stanford: On FlashAttention and sparsity, quantization, and efficient inference

Episode 32: Jamie Simon, UC Berkeley: On theoretical principles for how neural networks learn and generalize

Episode 31: Bill Thompson, UC Berkeley, on how cultural evolution shapes knowledge acquisition

Episode 30: Ben Eysenbach, CMU, on designing simpler and more principled RL algorithms

Jim Fan, NVIDIA: Foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant

Generally capable autonomous agents require embodiment to ground knowledge in a rich, interactive world, moving beyond the limitations of text-only foundation models. By leveraging massive, open-ended environments like Minecraft, researchers can train agents to solve complex, long-horizon tasks through natural language...

Sergey Levine, UC Berkeley: The bottlenecks to generalization in reinforcement learning, why simulation is doomed to succeed, and how to pick good research problems

Sergey Levine, a pioneer in modern deep reinforcement learning and professor at UC Berkeley, explores the transition from geometric-based robotic control to end-to-end deep learning systems. The conversation highlights the necessity of extreme design choices—such as the Arm Farm project—to isolate and validate the util...

Episode 27: Noam Brown, FAIR, on achieving human-level performance in poker and Diplomacy, and the power of spending compute at inference time

Episode 26: Sugandha Sharma, MIT, on biologically inspired neural architectures, how memories can be implemented, and control theory

Episode 25: Nicklas Hansen, UCSD, on long-horizon planning and why algorithms don't drive research progress