The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"

By Lukas Berglund, Meg Tong, Max Kaufmann, Mikita Balesni, Asa Cooper Stickland, Tomasz Korbak, Owain Evans. International Conference on Learning Representations, 2023

511 citations recorded by the source corpus.

We expose a surprising failure of generalization in auto-regressive large language models (LLMs). If a model is trained on a sentence of the form"A is B", it will not automatically generalize to the reverse direction"B is A". This is the Reversal Curse. For instance, if a model is trained on"Valentina Tereshkova was the first woman to travel to space", it will not automatically be able to answer the question,"Who was the first woman to travel to space?". Moreover, the likelihood of the correct answer ("Valentina Tershkova") will not be higher than for a random name. Thus, models do not generalize a prevalent pattern in their training set: if"A is B"occurs,"B is A"is more likely to occur. It is worth noting, however, that if"A is B"appears in-context, models can deduce the reverse relationship. We provide evidence for the Reversal Curse by finetuning GPT-3 and Llama-1 on fictitious statements such as"Uriah Hawthorne is the composer of Abyssal Melodies"and showing that they fail to correctly answer"Who composed Abyssal Melodies?". The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation. We also evaluate ChatGPT (GPT-3.5 and GPT-4) on questions about real-world celebrities, such as"Who is Tom Cruise's mother? [A: Mary Lee Pfeiffer]"and the reverse"Who is Mary Lee Pfeiffer's son?". GPT-4 correctly answers questions like the former 79% of the time, compared to 33% for the latter. Code available at: https://github.com/lukasberglund/reversal_curse.

View Paper: https://www.semanticscholar.org/paper/8eafec7014d08043517834b5a2ed26384f188873

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Fan Yang
2. Mengnan Du
3. Ninghao Liu
4. Haiyan Zhao
5. Adrian Weller
6. Zirui Zhao
7. Hanjie Chen
8. L. Yu

Citations in the corpus

• Large Language Models as Commonsense Knowledge for Large-Scale Task Planning (2023, background): “Even though LLMs are trained on internet-scale data, this strategy shows limits in generalization [10, 4], especially when encountering uncommon, complex tasks.”
• A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, unknown): “Furthermore, Berglund et al. (2023) unveiled a specific reasoning failure in LLMs termed the Reversal Curse .”
• Large Language Diffusion Models (2025, unknown): “• LLaDA effectively breaks the reversal curse [15] with consistent performance across forward and reversal tasks.”
• Explainability for Large Language Models: A Survey (2023, background): “Extensive experiments on multiple medical image datasets demonstrate the effectiveness and explainability of this training-free diagnostic pipeline.”
• Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data (2024, unknown): “From a probabilistic perspective, the sequential sampling process of auto-regressive models is inefficient and limits the reasoning ability in nonsequential orders (Berglund et al., 2023; Lv et al., 2023).”
• MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models (2023, result): “The Reversal Curse [5], where LLMs trained from a sentence “A is B” are not able to generalize to answer “B is A”, also aligns with the observation in this paper that LLMs lack backward mathematical reasoning ability.”