Encouraging general generalisation: Teaching Gemma (To Articulate) Why

·LessWrong··

TL;DRBuilding on previous research, I create a constitution for Gemma 3 27B-IT and attempt to Gemma to it in a way that generalises.I argue that alongside previously considered types of alignment generalisation (across dilemmas and across environments), a new type is worth considering: generalisation across principles (i.e. a model's ability to apply some part of its constitution helps it apply other parts). To evaluate if models' alignment generalises across principles, I hold out scenarios inv...

Read full article →

Related Articles

Opus 5.5 agents discover two room-temperature magnetic semiconductor candidates
outlier99 · Hacker News · 9h ago
Improper redaction reveals Google Data Center water and electricity usage
sensanaty · Hacker News · 1d ago
Pixel 11 doesn't yet meet the GrapheneOS security standards and may be skipped
finnlab · Hacker News · 17h ago
US closely monitoring case of lab worker who possibly died of plague in Siberia
tosh · Hacker News · 13h ago
Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
snehesht · Hacker News · 1d ago