⚡Megatron LLM Hub
←Back to search
Twitter

研究者认为“正则化器藏恶意代码”更像混淆手法而非可利用漏洞

llm供应链
This seems like a neat way to obfuscate some malicious code but I don’t see where it could be a vulnerability? By the time you’re running someone else’s regularizer you’re already running untrusted code and it could just tamper with the weights directly?