4 ms·
> While they provide data privacy, our research shows their weaker reasoning and alignment capabilities make them easier targets for sabotage. If you are using
by gok 1y ago
> While 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.
- liqilin1567 1y agoThis reminds me of stalwart's spam filter feature claim: "LLM-driven spam filtering and message analysis." :D https://github.com/stalwartlabs/stalwart https://github.com/stalwartlabs/stalwart