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

by   Lukas Berglund, et al.

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 "Olaf Scholz was the ninth Chancellor of Germany", it will not automatically be able to answer the question, "Who was the ninth Chancellor of Germany?". Moreover, the likelihood of the correct answer ("Olaf Scholz") will not be higher than for a random name. Thus, models exhibit a basic failure of logical deduction and do not generalize a prevalent pattern in their training set (i.e. if "A is B” occurs, "B is A" is more likely to occur). 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 latter. This shows a failure of logical deduction that we hypothesize is caused by the Reversal Curse. Code is available at


page 1

page 6

page 14

page 16


VQA-LOL: Visual Question Answering under the Lens of Logic

Logical connectives and their implications on the meaning of a natural l...

MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies

Autoregressive language models are trained by minimizing the cross-entro...

Learning to Generate Questions by Enhancing Text Generation with Sentence Selection

We introduce an approach for the answer-aware question generation proble...

Negated Complementary Commonsense using Large Language Models

Larger language models, such as GPT-3, have shown to be excellent in man...

SC-Ques: A Sentence Completion Question Dataset for English as a Second Language Learners

Sentence completion (SC) questions present a sentence with one or more b...

AutoQA: From Databases To QA Semantic Parsers With Only Synthetic Training Data

We propose AutoQA, a methodology and toolkit to generate semantic parser...

Learning Retrospective Knowledge with Reverse Reinforcement Learning

We present a Reverse Reinforcement Learning (Reverse RL) approach for re...

Please sign up or login with your details

Forgot password? Click here to reset