transformer vulnerabilities
1 post1 contributor
+ FollowBe the first to follow this
Write about itArticles about transformer vulnerabilities
Artificial Intelligence: How Hidden Triggers and Backdoors Compromise Transformer-Driven LLMs
LLMs like GPT-4 are vulnerable to stealthy backdoor attacks, from chain-of-thought hijacking to temporal triggers. The BackdoorLLM benchmark exposes these risks, highlighting the need for robust defenses. Learn about novel attack vectors like DarkMind and AdvBDGen, and defense strategies like CoS and SANDE to secure AI systems.
