Compositional prompt optimization with discrete codebooks
In brief
- PCO decomposes prompts into reusable units called natural instincts.
- Boosts performance by +13.5 points on Qwen3-8B and +11.8 on LLaMA-3.1-8B.
- Cuts deployed prompt length by up to 14.1x compared to MIPROv2.
A new automatic prompt optimization method for language models, PCO, breaks down instructions into reusable units. It significantly improves performance across benchmarks while reducing deployed prompt length.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error