Wasn't the shift you're describing also caused by realization that the west's 'marketplace of ideas' can be weaponized by foreign state actors, which do not subscribe to the ideals of freedom of expression?
At the same time those countries regulate the content that's available to their citizens very strictly.
I think that this openness combined with the very design of social media made performing active measures hugely more effective. I mean things like spreading disinformation, sewing discord, influencing elections, shaping public opinion...
For me, that's the biggest problem with the 'market of ideas' approach. I'm not arguing for or against it, but it's important to point out this.
Twitter has drastically scaled down moderation, and the result is that now anti-Semites, eugenics proponents and nazi-adjacent "race realists" are thriving.
It seems that platforms with lax moderation always end up attracting that crowd, so not sure "the marketplace of ideas" ever really existed to begin with.
Hmm… surely it’s a serious issue of freedom and liberty - most powerful tools are secretly censored, like removing books from library. Where are the libertarians here?
For completeness, it's probably worth noting that RLHF is not the whole picture for brand safety; for example, crude safety mechanisms are implemented through input filters, output filters, and hidden system prompts. That said, these are even less about any sort of a coherent moral compass; they're just band-aids to expediently fix problems without changing the model in any way.
Wasn't the shift you're describing also caused by realization that the west's 'marketplace of ideas' can be weaponized by foreign state actors, which do not subscribe to the ideals of freedom of expression?
At the same time those countries regulate the content that's available to their citizens very strictly.
I think that this openness combined with the very design of social media made performing active measures hugely more effective. I mean things like spreading disinformation, sewing discord, influencing elections, shaping public opinion...
For me, that's the biggest problem with the 'market of ideas' approach. I'm not arguing for or against it, but it's important to point out this.
Twitter has drastically scaled down moderation, and the result is that now anti-Semites, eugenics proponents and nazi-adjacent "race realists" are thriving.
It seems that platforms with lax moderation always end up attracting that crowd, so not sure "the marketplace of ideas" ever really existed to begin with.
RLHF is a whack-a-mole, we keep finding problematic behaviors and updating the training sets in response. Does it ever converge?
Hmm… surely it’s a serious issue of freedom and liberty - most powerful tools are secretly censored, like removing books from library. Where are the libertarians here?
For completeness, it's probably worth noting that RLHF is not the whole picture for brand safety; for example, crude safety mechanisms are implemented through input filters, output filters, and hidden system prompts. That said, these are even less about any sort of a coherent moral compass; they're just band-aids to expediently fix problems without changing the model in any way.