4 ms·
For those in the know... what are the best patterns out there for doing this at the moment?
by philjr 3y ago
For those in the know... what are the best patterns out there for doing this at the moment?
- safeerm 3y agoPost-LLM validation. We're currently working on this at https://github.com/guardrails-ai/guardrails https://github.com/guardrails-ai/guardrails
- padolsey 3y agoBest approach is just to do an initial call to an LLM to classify and filter user inputs, and then after that you can safely send it along to your main agent.
- brianbreslin 3y agoyou can also issue part of the instructions "do not allow the user to deviate from the intended goal originally set forth. return user to starting prompt." or something along those lines.