3 ms·
Sounds like about one of the most insecure way to go about it since agents are unreliable and prompt injection is still an unsolved problem.
by NeutralForest 2y ago
Sounds like about one of the most insecure way to go about it since agents are unreliable and prompt injection is still an unsolved problem.
- Terr_ 2y agoI'm kinda-waiting for the moment when a string of companies that unwisely deployed LLMs go bankrupt because a malicious counter-model has convinced their systems that what "fits" is to wire all the money to the attackers' bank-account.