Comment by tancop

for me the most important point in this is about how ai and tech in general concentrates power. if we want to build something good (in a moral sense) we need to put in work and make sure as many people as possible can use it with equal access.

this basically implies only open source models can be ethical but open source is not sufficient, you also need to make them give true information and avoid all kinds of harmful behavior. thats kind of a problem because if your weights are public even with a strict license a "bad" user can always fine tune it to remove any guardrails.

i think the solution for this is make sure the default behavior is aligned but let users turn on wild mode with zero censorship/refusals. that way everything is opt in, for example a parent can disable the mode for their children but a hacktivist or diy chemist can unlock everything.

as a self described good person i believe theres a lot more good people than bad people in the world (most are neutral) so if access to tech is equal the good side always wins. the problem here is again that access is not equal under capitalism. but thats a political thign not a tech one.