Why LLMs unfortunately won't be an anti-censorship tool
> Given this, a platform or government with a desire to censor could do it using another LLM to "review" the output of the first model and modify it according to the desired guidelines.
And even though people could use prompt engineering / ‘jailbreaks’ to circumvent this (get the LLM to phrase their response such that the censor won’t censor it), most people simply won’t take the trouble.
But what TC suggests is that that hobbles the usefulness of the responses to such an extent that China will be at a too severe disadvantage, losing too much relative economic growth, to keep it up.
Great to see you ported your content here and are talking about AI + China. It's one of my favorite thought experiments these days. Wild times.