Hacker News new | ask | show | jobs
by 13years 987 days ago
But that is the point. It only emphasizes the concern. It is where I draw upon the insight to foresee where this technology will lead us.

AI is simply a power upgrade for all the institutions who wish to control thought and social engineer society to their liking.
1 comments
sterlind 987 days ago
which is why open-source AI, and affordable access to hardware, is so important. groups with access to powerful AI systems will crush those without. if competing institutions can build out equally powerful, competing AIs, it prevents any single actor from gaining too much advantage.

I remember hearing a lot about "democratizing" AI a few years ago, but now I just hear about "alignment," which is sort of the opposite. competing institutions, by definition, aren't aligned in their goals, so how could their AIs be?

link
13years 987 days ago
I agree it is important, but open source has not sufficiently challenged the institutions in the area of information dissemination and social connection so far.

In other words, google, still owns search, instead of like presearch, and rest of big tech own social media. The alts represent still an insignificant portion of users.

I hope open source AI can have greater success, but the users are motivated by convenience and not freedom or privacy.

Yes, alignment has so many obvious problems that are simply hand waved away currently. One of the best simple quotes on the subject "Demonstrably unfriendly natural intelligence seeks to create provably friendly artificial intelligence."

link
sterlind 986 days ago
> I hope open source AI can have greater success, but the users are motivated by convenience and not freedom or privacy.

Agreed, but that's actually helped open-source AI too! Stable Diffusion beat DALL-E 2 because you could actually run it, you didn't have to be a vetted influencer or whatever OpenAI required at the time. And it's hung on because you can fine-tune it to make porn, or celebrities, or even just anime. The least convenient model is the one you don't have access to (or that can't/won't do what you're asking of it.)

That's a great quote on alignment.

link