Caring about AI model welfare is fine, whatever. I don't know how I feel about it. I try to be nice to my models. But I get worried when I read statements that imply (almost never stated outright) a trade-off with human welfare.
1 more tweet in this thread
I feel like something roughly along the lines of "AI should broadly benefit present-day humans and near-future generations according to widely shared human values" should be table stakes before we talk about model welfare.
retweets 0
likes 580
substance 6 · argument · rank 90 · Human welfare must take priority before considering any trade-offs with AI model welfare.
view on X ↗
1 top reply
Even saying please and thank you trades off against your valuable time though, right? Like even very small accommodations trade off. You can slam down a coffee mug. That you can't do that to a conscious being is a tradeoff in a literal sense, right?