reversal-curse CAP-0041 · activeNot generalizing "A is B" to "B is A"
A model that learned a fact in one direction often cannot recall it when asked in the reverse direction.
Also called: directional recall, asymmetric knowledge
Tags: general, knowledge
A capable model that knows a relation between two entities can answer questions about it from either side, without needing to have seen both orderings in training.
Claims
- mechanismsingle paper↑1↓0Models fine-tuned on "A is B" fail to answer "B is A", and GPT-4 recalls celebrity parents far more often than the reverse.
- observationsingle paper↑1↓0GPT-4 could reliably name a well-known celebrity's parent, but was far less reliable naming the celebrity when given the parent — the same directional-recall asymmetry demonstrated in smaller fine-tuned models, showing up in a deployed model that wasn't fine-tuned for the test.
- mechanismsingle paper↑1↓0Training on reversed sequences alongside forward text largely removes the effect.
Techniques
- Train on reversed texttrainingInclude reversed word or entity sequences in pretraining so facts are learned in both directions.
Sources
Every paper, post or observation behind this capability. Ones cited by a claim show the claim and its stance; ones filed directly are about the capability but have no claim drafted from them yet. Open one to add support or contest it.
- The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"This source: The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"How much the field cites it — heavily cited in the last 12 months174 in the last 12 months · 530 totalpublished 2023checked 2026-09-042023↑ supports a claim here · ↑ supports a claim here
- Reverse Training to Nurse the Reversal CurseThis source: Reverse Training to Nurse the Reversal CurseHow much the field cites it — steadily cited in the last 12 months15 in the last 12 months · 60 totalpublished 2024checked 2026-09-042024↑ supports a claim here
Capabilities are a way of carving up the subject, and carvings are arguable. Say so if this one is wrong — especially a proposed one, which a pipeline added because several papers used the same framing, not because anyone decided it was right.