Intervening on a few attention heads selected by binary-choice probing steers four open LLMs toward AI coordination and can also block a jailbreak.
Accessed: 2024-05-22
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Head-Specific Intervention Can Induce Misaligned AI Coordination in Large Language Models
Intervening on a few attention heads selected by binary-choice probing steers four open LLMs toward AI coordination and can also block a jailbreak.