Obvs you cannot hope that China would contribute positively to a regulatory effort, and you can't stop them producing models and handing them out. But if they are small enough to run locally, then that is pretty much the end of the game for regulating use. [Lots of people were doing this anyway].
It also makes it cheap enough for every nation to build its own LLM - therefore, the value moves to those who apply AI, not the LLM owners. Good for Europe.
Yep. Nation, company... On the value, yes, proportionately [as the value of developing the thing and selling it's use vanishes]. The value of applying it is the same as before the release.
Overall a win for both AI maximalism and the working class, as the implication is that AI can progress very quickly via open source and, in doing so, most of the relevant intellectual property will effectively be in the public domain, not owned by capitalists. (Also a third order win for the PRC.)
My personal evaluation:
- win for PRC: bad but 3rd order, mostly not a big deal
- win for working class: good because poverty is bad
- win for AI maximalism: good in the median outcome but perhaps bad when we fully account for low probability scenarios in the integral
Comments
- win for PRC: bad but 3rd order, mostly not a big deal
- win for working class: good because poverty is bad
- win for AI maximalism: good in the median outcome but perhaps bad when we fully account for low probability scenarios in the integral
HT: @george17.bsky.social