Comment on The security paradox of local LLMsComments−gok10moWhile they provide data privacy, our research shows their weaker reasoning and alignment capabilities make them easier targets for sabotage.If you are using any LLM's reasoning ability as a security boundary, something is deeply, deeply wrong.−liqilin156710moThis reminds me of stalwart's spam filter feature claim: "LLM-driven spam filtering and message analysis." :Dhttps://github.com/stalwartlabs/stalwart
Comments
If you are using any LLM's reasoning ability as a security boundary, something is deeply, deeply wrong.
This reminds me of stalwart's spam filter feature claim: "LLM-driven spam filtering and message analysis." :D
https://github.com/stalwartlabs/stalwart