用 Jev 与 CodeGraph 组合做安全审查
开发者分享了 typesafe-security-review 项目:把 Jev 的概率判断与代码图、OWASP/CWE 信息结合,用于低成本安全扫描。
作者阅读了Jev/KEV家族决策模型的论文,尝试验证是否能在这些模型中植入后门,如同普通LLM一样。
Not sure this is even possible...trying it anyway. Can you plant a backdoor in a calibrated decision model (the Jev/KEV family) the same way you can in a normal LLM? Spent the week reading about these models. Now I want to break one. paper: https://arxiv.org/html/2610.04985v1…