A new prompting technique called Consistency-based Self-adaptive Prompting (COSP) allows large language models to generate their own in-context examples, mitigating the drawbacks of traditional few-shot and zero-shot methods. Developed by researchers at Google, COSP leverages the model's own responses to create reliable examples, thereby improving reasoning accuracy without manual effort. The technique involves sampling multiple reasoning paths for a given query and using the consistency of the final answers as a quality filter, with low entropy indicating high confidence. AI
IMPACT This technique could improve LLM reasoning accuracy and reduce the manual effort required for prompt engineering.
RANK_REASON The cluster describes a new prompting technique detailed in a research paper. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
