One of the challenges on HN is that the United States pretty much singularly has a very strong right to free speech as a part of law. The right to be forgotten

One of the challenges on HN is that the United States pretty much singularly has a very strong right to free speech as a part of law. The right to be forgotten is a very direct an unarguable violation of this. It isn't even debatable, and thus we are left debating free speech issues of which Europeans and Americans have very strong differences (in my opinion, anyone who is under an authoritarian government doesn't get to participate in the debate for censorship.) I don't know what is worse - that, for example, a Norweigen neo-Nazi mass murderer who will only serve 20 years in prison could erase the stories about them and that event ever occuring, or that someone who did something really dumb in college could usurp and override everyone else's right to free speech. The right to be forgotten rests on the premise that someone can be "reformed" and that their right to erase the past outweighs the rights of everyone else. The basic premise is someone did something really bad or embarrassing and they are too lazy to change their name (I have friends who have done that, due to an embarrassing newspaper article appearing Google.) I think we are headed toward a bifurcated interest. The reason that countries outside of the US have been able to compel American tech companies to do their bidding, to date, is because these companies are monopolies that generate massive amounts of cashflow doing things which are not impacted by that law (right to be forgotten does almost 0 to Google's income.) How is this going to work with LLM models? Keyword censorship can be inserted at the input and output levels, but you can make the LLM bypass that. These models will not be re-trained on demand. The open models can not be recalled. I suspect a lot of countries are about to be excluded from the next wave of innovation.