Thousand-dimensional structure

·LessWrong··

Summary: One area we plan to explore at Resolution is personas and character training, operationalized as finding and controlling low-dimensional structure in models that emerges in pretraining and flows through post-training to superintelligence. The hope is to expand and systematize phenomena such as emergent misalignment, subliminal learning, and other empirical persona research, then intervene on this structure without accidentally hiding undesirable behavior elsewhere. If this approach reso...

Read full article →

Related Articles

GCC steering committee announces AI policy
arto · Hacker News · 4h ago
AI's top startups are barely publishing their research
YeGoblynQueenne · Hacker News · 18h ago
Why Is Everyone Trying to Build a Solid-State Battery?
crescit_eundo · Hacker News · 3h ago
Document-borne AI worms can self-propagate through Copilot for Word
Canopy9560 · Hacker News · 1d ago
Why Don't People Use Formal Methods? (2019)
Thom2503 · Hacker News · 3h ago