I personally think that in a decade or two the mainstream perspective on this will be similar to the outlook on cold-war era cryptography export controls: Long term futile, ineffective and burdensome, despite being "valuable" to adversaries.
I don't really see how locking the whole technology down is ever gonna work (unless hardware progress magically stops)-- I'd expect effort/HW-requirement vs. "intelligence/capability" curves to look somewhat like a logistics function (exponential at first with quick growth-- think GPT3 era, then leveling off to a sub-linear regime where improvements requires exponentially more hardware/training).
As long as the underlying hardware improves, we'll invariably end up with Mythos like capabilities that can be achieved with consumer hardware within a decade or so at the latest...
I'm grateful to live in such interesting times, though.
I personally think that in a decade or two the mainstream perspective on this will be similar to the outlook on cold-war era cryptography export controls: Long term futile, ineffective and burdensome, despite being "valuable" to adversaries.
I don't really see how locking the whole technology down is ever gonna work (unless hardware progress magically stops)-- I'd expect effort/HW-requirement vs. "intelligence/capability" curves to look somewhat like a logistics function (exponential at first with quick growth-- think GPT3 era, then leveling off to a sub-linear regime where improvements requires exponentially more hardware/training).
As long as the underlying hardware improves, we'll invariably end up with Mythos like capabilities that can be achieved with consumer hardware within a decade or so at the latest...
I'm grateful to live in such interesting times, though.
Discussions: https://news.ycombinator.com/item?id=48678789
https://news.ycombinator.com/item?id=48683021
and official: https://news.ycombinator.com/item?id=48689028
Thanks for the links! I somehow missed those topic completely and failed to find them before posting, too.