Output Validation

Output validators assess the LLM-generated outputs to detect logical inconsistencies, hallucinations, and potential biases. This ensures that the LLM responses are accurate, reliable, and unbiased.

Pythia output validators serve the following objectives to achieve reliable AI:

• Implement checks against a predefined set of rules and patterns.

• Flag outputs that seem illogical, inconsistent, or biased.

• Provide detailed logs and alerts for flagged outputs.

• 
  

Pythia implements the following output validators in each API request:

• Detect PII

• Gibberish Text

• Profanity Free

• Toxic Language

• QA Relevance LLM

• Saliency Check

• Secrets Present

• Financial Tone