This claim strikes me as false: >”Securing model weights is therefore the most critical factor for maintaining the U.S. AI advantage.” Model weights are ephem

This claim strikes me as false: >”Securing model weights is therefore the most critical factor for maintaining the U.S. AI advantage.” Model weights are ephemera. What is NOT ephemera is work on embodied AI systems that live and learn in the “real” world. Embodied recurrent learning system will probably be harbingers of AGI or the real deal itself. There are already a large number of potential “real” worlds and cultures for embodied AI systems to grow up in. They will almost inevitably assume characteristics of the culture(s) in which they originate. Will they be “raised” in an US capitalist culture or a Chinese Communist Party culture? A Norwegian culture or a Saudi culture? Will their philosophical fundamentals lean toward X, Y, or Z? AGI systems will replicate the diversity of cultures and I see no strong logic to assume that this will either precipitate conflict or reduce it. Like all technological advances it is an accelerant of cultural trends. For the SciFi version of where this could lead see Hannu Rajaniemi’s The Quantum Thief trilogy: https://en.wikipedia.org/wiki/The_Quantum_Thief https://en.wikipedia.org/wiki/The_Quantum_Thief