AI alignment has two sides: machines and people. Computer scientists study the first, social scientists the second, and we need more collaboration between these spheres.

Interesting finding by @danicajdillion.bsky.social et al. - a good reminder that how we connect matters just as much as what we believe. #SocialScience #SocialNetwork #USPol

We are facing a crisis of disconnection and this is directly linked to polarization. New research finds that people living in communities with denser social networks consistently reported less partisan animosity and a greater sense of national unity. link.springer.com/epdf/10.1007...

Americans are growing more socially isolated and politically divided. Our new paper in Applied Network Science suggests these two forms of disconnection may be linked. People with denser, more connected social networks often feel less partisan animosity. 🧵

A person and a puppy are drowning in a lake. You can only save one—who do you choose? 1 in 5 dog owners would choose the puppy. Are our relationships with our fur babies getting in the way of those with people? New post (🔗 below 👇)

🌍 Introducing WorldValuesBench! A benchmark to evaluate how well LLMs reflect cultural differences in human values. Built from 94k+ participants in the World Values Survey → 20M examples of (demographics, value question → answer). 🧵

1 in 5 dog owners would save a puppy's life over yours. 🐶 The majority of pet owners see their dog as a soulmate. Many don’t just prefer their own dog over people—they’ll pick a stranger’s dog over a human life. New preprint 🧵

New preprint 🚨 Cognitive bottlenecks make LLMs more morally aligned with people 🧠🤖 We made AI “think” more like people by narrowing its focus to a few key moral cues. This AI better predicted people’s moral judgments & was more trusted. 🧵 ⬇️