Language models can learn implicit multi-hop reasoning, but only if they have lots of training data

2025/05/23 by Yao, Yuekun, Du, Yupei, Zhu, Dawei +2 · 2 citations
#Computation and Language (cs.CL) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2505.17923

Abstract

Implicit reasoning is the ability of a language model to solve multi-hop reasoning tasks in a single forward pass, without chain of thought. We investigate this capability using GPT2-style language models trained from scratch on controlled k-hop reasoning datasets (k = 2, 3, 4). We show that while such models can indeed learn implicit k-hop reasoning, the required training data grows exponentially in k, and the required number of transformer layers grows linearly in k. We offer a theoretical explanation for why this depth growth is necessary. We further find that the data requirement can be mitigated, but not eliminated, through curriculum learning.

Cited by

• Benefits and Limitations of Communication in Multi-Agent Reasoning
• Rote Learning Considered Useful: Generalizing over Memorized Data in LLMs

Related

• How do Transformers Learn Implicit Reasoning?
• Implicit Reasoning in Transformers is Reasoning through Shortcuts
• Large Language Models Still Face Challenges in Multi-Hop Reasoning with External Knowledge
• Do Larger Language Models Generalize Better? A Scaling Law for Implicit Reasoning at Pretraining Time
• Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge