Implicit Reasoning in Transformers is Reasoning through Shortcuts

• URL: http://arxiv.org/abs/2503.07604v2
• Date: Tue, 18 Mar 2025 12:08:17 GMT
• Title: Implicit Reasoning in Transformers is Reasoning through Shortcuts
• Authors: Tianhe Lin, Jian Xie, Siyu Yuan, Deqing Yang,
• Abstract summary: Test-time compute is emerging as a new paradigm for enhancing language models' complex multi-step reasoning capabilities.<n>We investigate how language models perform implicit reasoning in multi-step tasks.
• Score: 10.351525484558376
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Test-time compute is emerging as a new paradigm for enhancing language models' complex multi-step reasoning capabilities, as demonstrated by the success of OpenAI's o1 and o3, as well as DeepSeek's R1. Compared to explicit reasoning in test-time compute, implicit reasoning is more inference-efficient, requiring fewer generated tokens. However, why does the advanced reasoning capability fail to emerge in the implicit reasoning style? In this work, we train GPT-2 from scratch on a curated multi-step mathematical reasoning dataset and conduct analytical experiments to investigate how language models perform implicit reasoning in multi-step tasks. Our findings reveal: 1) Language models can perform step-by-step reasoning and achieve high accuracy in both in-domain and out-of-domain tests via implicit reasoning. However, this capability only emerges when trained on fixed-pattern data. 2) Conversely, implicit reasoning abilities emerging from training on unfixed-pattern data tend to overfit a specific pattern and fail to generalize further. Notably, this limitation is also observed in state-of-the-art large language models. These findings suggest that language models acquire implicit reasoning through shortcut learning, enabling strong performance on tasks with similar patterns while lacking generalization.

Related papers

• MastermindEval: A Simple But Scalable Reasoning Benchmark [3.5519847710183674]
  We introduce MastermindEval, a simple, scalable, and interpretable deductive reasoning benchmark inspired by the board game Mastermind. Our benchmark supports two evaluation paradigms: (1) agentic evaluation, in which the model autonomously plays the game, and (2) deductive reasoning evaluation, in which the model is given a pre-played game state with only one possible valid code to infer.
  arXiv  Detail & Related papers  (2025-03-07T19:24:59Z)
• Unveiling Reasoning Thresholds in Language Models: Scaling, Fine-Tuning, and Interpretability through Attention Maps [3.8936716676293917]
  This study investigates the in-context learning capabilities of various decoder-only transformer-based language models with different model sizes and training data.<n>We identify a critical parameter threshold (1.6 billion), beyond which reasoning performance improves significantly in tasks such as commonsense reasoning in multiple-choice question answering and deductive reasoning.
  arXiv  Detail & Related papers  (2025-02-21T00:48:32Z)
• Improving Arithmetic Reasoning Ability of Large Language Models through Relation Tuples, Verification and Dynamic Feedback [14.938401898546553]
  We propose to use a semi-structured form to represent reasoning steps of large language models. Specifically, we use relations, which are not only human but also machine-friendly and easier to verify than natural language.
  arXiv  Detail & Related papers  (2024-06-25T18:21:00Z)
• Distilling Reasoning Ability from Large Language Models with Adaptive Thinking [54.047761094420174]
  Chain of thought finetuning (cot-finetuning) aims to endow small language models (SLM) with reasoning ability to improve their performance towards specific tasks. Most existing cot-finetuning methods adopt a pre-thinking mechanism, allowing the SLM to generate a rationale before providing an answer. This mechanism enables SLM to analyze and think about complex questions, but it also makes answer correctness highly sensitive to minor errors in rationale. We propose a robust post-thinking mechanism to generate answers before rationale.
  arXiv  Detail & Related papers  (2024-04-14T07:19:27Z)
• Transformers in the Service of Description Logic-based Contexts [2.8210912543324658]
  We construct the natural language dataset, DELTA$_D$, using the description logic language $mathcalALCQ$. We investigate the reasoning ability of a supervised fine-tuned DeBERTa-based model and of two large language models (GPT-3.5, GPT-4) with few-shot prompting. Our results demonstrate that the DeBERTa-based model can master the reasoning task and that the performance of GPTs can improve significantly even when a small number of samples is provided.
  arXiv  Detail & Related papers  (2023-11-15T13:23:24Z)
• UNcommonsense Reasoning: Abductive Reasoning about Uncommon Situations [62.71847873326847]
  We investigate the ability to model unusual, unexpected, and unlikely situations. Given a piece of context with an unexpected outcome, this task requires reasoning abductively to generate an explanation. We release a new English language corpus called UNcommonsense.
  arXiv  Detail & Related papers  (2023-11-14T19:00:55Z)
• Implicit Chain of Thought Reasoning via Knowledge Distillation [58.80851216530288]
  Instead of explicitly producing the chain of thought reasoning steps, we use the language model's internal hidden states to perform implicit reasoning. We find that this approach enables solving tasks previously not solvable without explicit chain-of-thought, at a speed comparable to no chain-of-thought.
  arXiv  Detail & Related papers  (2023-11-02T17:59:49Z)
• Interpretability at Scale: Identifying Causal Mechanisms in Alpaca [62.65877150123775]
  We use Boundless DAS to efficiently search for interpretable causal structure in large language models while they follow instructions. Our findings mark a first step toward faithfully understanding the inner-workings of our ever-growing and most widely deployed language models.
  arXiv  Detail & Related papers  (2023-05-15T17:15:40Z)
• APOLLO: A Simple Approach for Adaptive Pretraining of Language Models for Logical Reasoning [73.3035118224719]
  We propose APOLLO, an adaptively pretrained language model that has improved logical reasoning abilities. APOLLO performs comparably on ReClor and outperforms baselines on LogiQA.
  arXiv  Detail & Related papers  (2022-12-19T07:40:02Z)
• ALERT: Adapting Language Models to Reasoning Tasks [43.8679673685468]
  ALERT is a benchmark and suite of analyses for assessing language models' reasoning ability. ALERT provides a test bed to asses any language model on fine-grained reasoning skills. We find that language models learn more reasoning skills during finetuning stage compared to pretraining state.
  arXiv  Detail & Related papers  (2022-12-16T05:15:41Z)
• MetaLogic: Logical Reasoning Explanations with Fine-Grained Structure [129.8481568648651]
  We propose a benchmark to investigate models' logical reasoning capabilities in complex real-life scenarios. Based on the multi-hop chain of reasoning, the explanation form includes three main components. We evaluate the current best models' performance on this new explanation form.
  arXiv  Detail & Related papers  (2022-10-22T16:01:13Z)
• Making Large Language Models Better Reasoners with Step-Aware Verifier [49.16750018427259]
  DIVERSE (Diverse Verifier on Reasoning Step) is a novel approach that further enhances the reasoning capability of language models. We evaluate DIVERSE on the latest language model code-davinci and show that it achieves new state-of-the-art results on six of eight reasoning benchmarks.
  arXiv  Detail & Related papers  (2022-06-06T03:38:36Z)
• Does entity abstraction help generative Transformers reason? [8.159805544989359]
  We study the utility of incorporating entity type abstractions into pre-trained Transformers. We test these methods on four NLP tasks requiring different forms of logical reasoning.
  arXiv  Detail & Related papers  (2022-01-05T19:00:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.