Blog
Dados & Embeddings
Self-Recognition Finetuning can Prevent and Reverse Emergent Misalignment
arXiv:2606.23700v1 Announce Type: new Abstract: Emergent misalignment (EM) has been linked to the activation of misaligned persona vectors and evil character traits, suggesting that EM operates through disruption of the model's aligned character rather than direct learning of harmful content. Motivated by this connection, we study self-generated text recognition (SGTR) finetuning as a character-targeted intervention that is distinct from existing in-training defenses. We conduct two-stage finetu...
arXiv cs.CL
·Arush Tagade, Shaoheng Zhou, Jiaxin Wen, Shi Feng
·
// relacionados
Leia também
Blog
Unicorn, pelican, Middle-earth: OpenAI co-founder Karpathy is looking for the next AI vibe test
Editorial
CAPA: o benchmark que mede se o assistente de código aprende com você — ou repete a mesma pergunta
Blog
Why biological data matters more in AI drug discovery
Blog