zlacker

Alignment is considered an extremely hard problem for a reason. It's already nigh impossible when you're dealing with humans.

Btw: do you think ridicule eould be helpful here?

>>concor+(OP)
I can see how ridicule of this specific instance could be the best medicine for an optimal outcome, even by a utilitarian argument, which I generally don't like to make by the way. It is indeed nigh impossible, which is kind of my point. They could have shown more humility. If anything, this whole debacle has been a moral victory for e/acc, seeing how the brightest of minds are at a loss dealing with alignment anyway.

>>arkety+Q
I don't understand how the conclusion of this is "so we should proceed with AI" rather than "so we should immediately outlaw all foundation model training". Clearly corporate self-governance has failed completely.