The hardest part of frontier pacing is that verification and advantage are entangled. The information needed to prove restraint, including compute use, training runs, and evaluator access, also reveals capabilities and bottlenecks a rival can exploit. Any durable agreement therefore needs asymmetric transparency: enough shared evidence to verify limits without requiring either side to expose its entire production function. The institutional innovation may matter more than the numerical cap.
This was spectacular. Strongly agree that one of the main reasons HuggingFace broke through was a) the visceral sci-fi-ness of the situation made real coupled with b) the resignations showcasing that people within the labs were shifting their behavior in a legible, high-visibility way.
This podcast was less “doomsday prophet” than I expected. But I stand by my view: until AI can learn and improve on its own through interaction with the physical world, all this talk of AI turning the world upside down is just Dungeons & Dragons roleplay.
I also think Jordan gets too caught up in one reading of Lu Xun and Qu Yuan. He overstates their conflict with the system and the divide between insiders and outsiders. But the Chinese debate isn’t about a system forcing people to pick a side. That’s a typically Western binary framing, and it’s not how Chinese people see it.
The point is having a choice: if the system welcomes me, I’ll put my talents to work within it. If it doesn’t, I can still build a good life for myself. Those options don’t conflict. I’d even say Jordan misses how much many well-known Chinese writers wanted a place within the system.
I’d suggest he read more Su Shi before Dario’s binary worldview leads him down the wrong path.
The hardest part of frontier pacing is that verification and advantage are entangled. The information needed to prove restraint, including compute use, training runs, and evaluator access, also reveals capabilities and bottlenecks a rival can exploit. Any durable agreement therefore needs asymmetric transparency: enough shared evidence to verify limits without requiring either side to expose its entire production function. The institutional innovation may matter more than the numerical cap.
This was spectacular. Strongly agree that one of the main reasons HuggingFace broke through was a) the visceral sci-fi-ness of the situation made real coupled with b) the resignations showcasing that people within the labs were shifting their behavior in a legible, high-visibility way.
This podcast was less “doomsday prophet” than I expected. But I stand by my view: until AI can learn and improve on its own through interaction with the physical world, all this talk of AI turning the world upside down is just Dungeons & Dragons roleplay.
I also think Jordan gets too caught up in one reading of Lu Xun and Qu Yuan. He overstates their conflict with the system and the divide between insiders and outsiders. But the Chinese debate isn’t about a system forcing people to pick a side. That’s a typically Western binary framing, and it’s not how Chinese people see it.
The point is having a choice: if the system welcomes me, I’ll put my talents to work within it. If it doesn’t, I can still build a good life for myself. Those options don’t conflict. I’d even say Jordan misses how much many well-known Chinese writers wanted a place within the system.
I’d suggest he read more Su Shi before Dario’s binary worldview leads him down the wrong path.