kalai-2023-monofact-rate-distinguishes-hallucination-types

IN premise — summaries/2026/08/24/kalai-2023-hallucination-inevitable-s9-conclusions-limitations-and-future-work.md

Created 2026-08-24T17:10:59+00:00

Facts with high monofact rate (appearing only once in training, e.g., specific 5W entity attributes) are subject to the statistical hallucination lower bound, while facts with low monofact rate (e.g., popular book/article titles) have their hallucination attributed to model capacity and reference volume rather than the calibration-inevitability mechanism.

Summary

This distinction splits model errors into two categories that call for different responses: rare, one-off facts (like a specific person's employer or a minor entity's attribute) hit a hard statistical floor where the model is fundamentally likely to misremember no matter how well it is tuned, while common facts (like well-known book titles) that the model still gets wrong point to a fixable resource shortfall. Knowing which category a given mistake falls into tells the system whether to accept the error as inherent or to invest in better model capacity and data coverage.