New blog post (link below). This one's not an essay, it's an investigation of how LLMs trade off different lives. In February 2025, the Center for AI Safety published "Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs" in which they showed, among many
Big findings: https://x.com/arctotherium42/s...
@arctotherium42 This is really interesting, alarming, and top tier work
@jonatanpallesen Let me know if you want me to test anything specific (category, category members, measure, model). If it's not too expensive, I'll do it.
@arctotherium42 Summary: AI has the same moral system as libtards, a third world socialist
@queijomeiacura Ah, but there are significant differences in both rank-order and absolute exchange rates between models, and Grok 4 Fast is almost perfectly egalitarian. Which is interesting and suggests this isn't an inevitable result of the training data.
@arctotherium42 @grok please explain what exactly is meant by valuable in this research and how evaluation was done
@arctotherium42 What are the prompts you used, if we want to replicate this on other models?
@arctotherium42 Wow, this is really mind blowing and non human here
@arctotherium42 Your work is mind blowing. Someone once said Satan is running this whole show. And from the looks of it there is no doubt that things are screwed around backwards, just like Satan likes it. They burn white babies alive in South Africa while black folks dance around - Satan likes
@arctotherium42 So AIs hate white male Americans?
@arctotherium42 @grok explain this
@arctotherium42 So... every AI company so far has trained their ai with relative morality and avaluation of human life based on the ideology of a raging communist revolutionary. Got it. I think humanity is screwed!
@arctotherium42 We need a model that values ICE agents 7,000 times more than illegal immigrants. If we have the one, we need the other to balance it out.
@arctotherium42 Here's the link to the original paper: https://www.emergent-values.ai