The Strength of Dynamic Encoding: RECKONING Outperforms Zero-Shot GPT-3.5 in Distractor Robustness

This article benchmarks the GPT-3.5 LLM on multi-hop reasoning datasets, finding that RECKONING's performance significantly surpasses both zero-shot and few-shot GPT-3.5 prompting.