Automatic Prompt Optimization
A technical direction that automatically searches for or rewrites prompt text through algorithms to maximize model performance on target tasks without modifying model weights. Common approaches include soft prompt optimization, text-level optimization, and reinforcement learning-driven policy search.
Timeline (last 90 days)
Automatic Prompt Optimization (APO) automatically searches for or rewrites prompt text through algorithms to improve model performance on target tasks without adjusting model weights.
Most existing structured prompt optimization methods are general-purpose designs and do not specifically target the semantic characteristics of stance labels and the category-specific ambiguities.
Common APO methods include soft prompt optimization, text-level optimization based on LLM generation capabilities (such as OPRO and ProTeGi), and reinforcement learning-driven policy search.
All Facts (3)
Automatic Prompt Optimization (APO) automatically searches for or rewrites prompt text through algorithms to improve model performance on target tasks without adjusting model weights.
50%UnverifiedMost existing structured prompt optimization methods are general-purpose designs and do not specifically target the semantic characteristics of stance labels and the category-specific ambiguities.
50%UnverifiedCommon APO methods include soft prompt optimization, text-level optimization based on LLM generation capabilities (such as OPRO and ProTeGi), and reinforcement learning-driven policy search.
50%