AAIA: Advanced in AI Audit

Reducing top-p sampling limits generation to a narrower set of more probable next tokens, making model output less stochastic and reducing hallucinated or fabricated content. Increasing temperature generally makes responses more variable, while added context and frequency penalties do not directly control sampling randomness.