reversal-curse-shows-up-in-a-deployed-model-gpt-4 CLM-0137observationsingle paperReviewed by AI and Russ Miller
GPT-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.
Capability: Not generalizing "A is B" to "B is A"
Observed on
GPT. 2023, GPT-4 class. General.
Model cards: GPT-4
Sources
- GPT-4 named a celebrity's parent correctly far more often than it could name the celebrity when given the parent.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- Models fine-tuned on "A is B" fail to answer "B is A", and GPT-4 recalls celebrity parents far more often than the reverse.Not generalizing "A is B" to "B is A"
- As a raw pairwise judge with no order-swap mitigation, GPT-4 can have its verdict flipped simply by swapping which candidate answer is shown first; applying a position-swap protocol restores agreement with careful human raters to close to human-human agreement levels.Biased when judging other outputs
- GPT-4 recommended non-existent Python and JavaScript packages in only a small percentage of generations — lower than the open models tested in the same study — but the same hallucinated names recurred often enough across runs to be practically exploitable by an attacker who registers them ahead of time.Writing secure code and dependencies
- On 430 tasks distilled from real professional-forum threads across seven domains, the best model (GPT-5) scores about 54%, under the 60% passing line the authors set, with computer science and applied technology lowest at roughly 40-45%, so long-tail practitioner knowledge remains a gap even where standard benchmarks are saturated.Competence in a specialist domain · reviewed by AI
- For 1.5B-8B base models given a domain SFT expert (math or medical), keeping both models and routing token by token — invoking the expert only on tokens where a trained router predicts the base would fail — preserves general-benchmark accuracy within about 0.5% of the base while exceeding the expert's in-domain accuracy, whereas loss-modifying SFT regularizers still lose up to 16% general accuracy.Retaining skills and safety after fine-tuning · reviewed by AI