hop-depth-decreases-accuracy-all-methods
IN premise — summaries/2026/08/24/zhong-2023-mquake-sR-references-chunk-2.md
Created 2026-08-25T02:59:08+00:00
Across all editing methods tested (Base, FT, MEND, ROME, MEMIT), 2-hop accuracy is substantially higher than 3-hop, and 3-hop higher than 4-hop, indicating multi-hop depth is the primary performance differentiator.
Summary
No matter which fact-editing technique you use, the number of reasoning steps a query requires is what really determines success, with each additional hop in the chain causing a significant accuracy drop. This means the system should prioritize handling deeper multi-hop chains as the main bottleneck rather than swapping between editing algorithms, since depth is the primary performance driver across every method tested.