Controlling LLM Output

With Prediction Guard, you can control the output of LLMs in the following ways (which can be mixed and matched according to your use case and constraints):

• Enforce output structure and/or types
• Check for consistency
• Filter for factuality and toxicity