Yeah, that's gonna be a tough one. Do you pick the side of open models you can run and tune in your own country to have control over them, or do you rely on American models that can be turned off at a whim. Seems like the choice is pretty obvious for any country that has even a modicum of sovereignty.
But also, this moves so fast it's only 80 days between Mythos and GLM-5.3*, so the former getting blocked even for just 18 days wiped out 22.5% of the potential head-start advantage of picking the US.
That said, article seems to say the USA's Pax Silica** is about manufacturing the hardware, while China's alternative is about software.
A strange game. I think China's going to win a battle over hardware and supply chains; while a few years ago I'd have expected the USA to win a battle over software, less so today and this software in particular.
* people I trust seem worried about the cyberattack potential of GLM-5.3 in same way as Mythos; I appreciate some may roll their eyes given how often somone says something like that about different AI models, but remember I'm talking abouy catch-up delay here, not absolute skill.
** The two hard problems in politics are naming things and…
Yeah, that's gonna be a tough one. Do you pick the side of open models you can run and tune in your own country to have control over them, or do you rely on American models that can be turned off at a whim. Seems like the choice is pretty obvious for any country that has even a modicum of sovereignty.
Indeed.
But also, this moves so fast it's only 80 days between Mythos and GLM-5.3*, so the former getting blocked even for just 18 days wiped out 22.5% of the potential head-start advantage of picking the US.
That said, article seems to say the USA's Pax Silica** is about manufacturing the hardware, while China's alternative is about software.
A strange game. I think China's going to win a battle over hardware and supply chains; while a few years ago I'd have expected the USA to win a battle over software, less so today and this software in particular.
* people I trust seem worried about the cyberattack potential of GLM-5.3 in same way as Mythos; I appreciate some may roll their eyes given how often somone says something like that about different AI models, but remember I'm talking abouy catch-up delay here, not absolute skill.
** The two hard problems in politics are naming things and…