adaptive-retrieval-cost-savings-15pct-entityquestions

IN premise — summaries/2026/08/24/mallen-2023-when-not-to-trust-s7-discussion-and-conclusions.md

Created 2026-08-25T02:58:12+00:00

Adaptive Retrieval cuts GPT-3 davinci 003 API costs by approximately 15% on EntityQuestions while matching retrieval-only accuracy

Summary

For entity-type questions, the adaptive retrieval approach spends about 15% less on GPT-3 davinci-003 API calls compared to a plain retrieval-only setup, without any measurable drop in answer quality. In practice, this means the system can route those question types through adaptive retrieval and save money on every call while still getting the same correct answers it would from the simpler pipeline.