Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding
Paper • 2606.21906 • Published • 20 •
12
Darius Farnunkle
vmlinux
·
AI & ML interests
Goofy stuff
Recent Activity
commentedon a paper 1 day ago
Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer DecodingOrganizations
None yet
commented a
paper 1 day ago