Cover Image

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

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Berglund, Lukas, Tong, Meg, Kaufmann, Max, Balesni, Mikita, Stickland, Asa Cooper, Korbak, Tomasz, Evans, Owain
Format:	Preprint
Published:	2023
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Lessons from Studying Two-Hop Latent Reasoning
by: Balesni, Mikita, et al.
Published: (2024)

How to evaluate control measures for LLM agents? A trajectory from today to superintelligence
by: Korbak, Tomek, et al.
Published: (2025)

Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs
by: Laine, Rudolf, et al.
Published: (2024)

Negation Neglect: When models fail to learn negations in training
by: Mayne, Harry, et al.
Published: (2026)

Large Language Models can Strategically Deceive their Users when Put Under Pressure
by: Scheurer, Jérémy, et al.
Published: (2023)

Does Unlearning Truly Unlearn? A Black Box Evaluation of LLM Unlearning Methods
by: Doshi, Jai, et al.
Published: (2024)

Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs
by: Price, Sara, et al.
Published: (2024)

Reverse Training to Nurse the Reversal Curse
by: Golovneva, Olga, et al.
Published: (2024)

Aligning language models with human preferences
by: Korbak, Tomasz
Published: (2024)

An Analysis and Mitigation of the Reversal Curse
by: Lv, Ang, et al.
Published: (2023)

Tell me about yourself: LLMs are aware of their learned behaviors
by: Betley, Jan, et al.
Published: (2025)

Looking Inward: Language Models Can Learn About Themselves by Introspection
by: Binder, Felix J, et al.
Published: (2024)

The Factorization Curse: Which Tokens You Predict Underlie the Reversal Curse and More
by: Kitouni, Ouail, et al.
Published: (2024)

Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs
by: Sheshadri, Abhay, et al.
Published: (2024)

The Illusion of Latent Generalization: Bi-directionality and the Reversal Curse
by: Coda-Forno, Julian, et al.
Published: (2026)

Catalytic Role Of Noise And Necessity Of Inductive Biases In The Emergence Of Compositional Communication
by: Kuciński, Łukasz, et al.
Published: (2021)

A Theoretical Analysis of Why Masked Diffusion Models Mitigate the Reversal Curse
by: Jeon, Moongyu, et al.
Published: (2026)

Why Do Language Model Agents Whistleblow?
by: Agrawal, Kushal, et al.
Published: (2025)

Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data
by: Gerstgrasser, Matthias, et al.
Published: (2024)

Steering Without Side Effects: Improving Post-Deployment Control of Language Models
by: Stickland, Asa Cooper, et al.
Published: (2024)

Thought Crime: Backdoors and Emergent Misalignment in Reasoning Models
by: Chua, James, et al.
Published: (2025)

Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data
by: Treutlein, Johannes, et al.
Published: (2024)

Towards evaluations-based safety cases for AI scheming
by: Balesni, Mikita, et al.
Published: (2024)

Mitigating Reversal Curse in Large Language Models via Semantic-aware Permutation Training
by: Guo, Qingyan, et al.
Published: (2024)

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
by: Betley, Jan, et al.
Published: (2025)

Exploring the Reversal Curse and Other Deductive Logical Reasoning in BERT and GPT-Based Large Language Models
by: Wu, Da, et al.
Published: (2023)

Frontier Models are Capable of In-context Scheming
by: Meinke, Alexander, et al.
Published: (2024)

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
by: Betley, Jan, et al.
Published: (2025)

Evaluating the Reversal Curse in Model Editing
by: Xu, Hao-Xiang, et al.
Published: (2023)

When Does Sparsity Mitigate the Curse of Depth in LLMs
by: Muhtar, Dilxat, et al.
Published: (2026)

Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
by: McKee-Reid, Leo, et al.
Published: (2024)

Training Language Models with Language Feedback at Scale
by: Scheurer, Jérémy, et al.
Published: (2023)

Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers
by: Karvonen, Adam, et al.
Published: (2025)

Time-Reversal Provides Unsupervised Feedback to LLMs
by: Varun, Yerram, et al.
Published: (2024)

No Need for Explanations: LLMs can implicitly learn from mistakes in-context
by: Alazraki, Lisa, et al.
Published: (2025)

Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions
by: Li, Xiang, et al.
Published: (2024)

Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs
by: Lin, Haokun, et al.
Published: (2025)

Induction Head Toxicity Mechanistically Explains Repetition Curse in Large Language Models
by: Wang, Shuxun, et al.
Published: (2025)

Adaptive Layer-skipping in Pre-trained LLMs
by: Luo, Xuan, et al.
Published: (2025)

Layerwise Recall and the Geometry of Interwoven Knowledge in LLMs
by: Lei, Ge, et al.
Published: (2025)