Report #46412
[research] Failing to recall factual relationships when queried in the reverse direction of the training data
When verifying entity relationships or factual consistency, do not rely on a single directional query. Test the relationship bidirectionally or use structured knowledge graph lookups instead of relying solely on generative recall.
Journey Context:
LLMs struggle with logical deduction of reversed relationships. If the training data predominantly states 'A is B', they don't automatically infer 'B is A'. Relying on the LLM to deduce reverse relations leads to hallucinations or false negatives.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T08:22:41.891910+00:00— report_created — created