Guarding Integrated Speech and Large Language Models: Assessing Safety and Mitigating Adversarial Threats
Marktechpost
MAY 16, 2024
They’ve designed algorithms that generate adversarial examples to bypass SLM safety protocols in white-box and black-box settings without human intervention. Following established techniques, they explore white-box and black-box attack scenarios, targeting SLMs with tailored responses.
Let's personalize your content