↳
In-reply-to
»
A very insightful chat tonight with @prologic, @ocdtrekkie, @darch, and @ocdtrekkie's friend Ian who is, presumably, still primarily using legacy social networks. Hopefully we can change that. Some things we talked about tonight:
⤋ Read More
@mckinley@twtxt.net Hmm yeah I saw something about this in my feed the other day. The problem is I’m not even sure how you’d protect against such attacks. Because the GPT-3 prompts themselves are basically just English which as we know has quite a wide variation of structure, it would be next to impossible to detect Prompt Injection attacks on the input. Sanitizing input for SQL is a much easier task as you are dealing with a stricter subset of input and a language with a pre-defined grammar.