Dwarkesh Podcast

Adam Marblestone × Dwarkesh Patel

Adam Marblestone’s differentiated hypothesis is that evolution encoded many stage-specific loss functions and developmental curricula, helping explain how a 3-gigabyte genome can produce sample-efficient intelligence.The investable timeline is roughly 10 years, with a transformative event still more than five years away; meanwhile, low-billions neuroscience funding and cheaper connectomics could test the thesis against trillions in GPU capex, while reward machinery remains an alignment risk.