Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning

2022/05/19 by Antonia Creswell, Murray Shanahan, Creswell, Antonia +3 · 1 voice · 53 citations
Computer Science · #Explainable Artificial Intelligence (XAI) #Natural Language Processing Techniques #Topic Modeling #cs.AI #cs.CL

paper · pdf · doi:10.48550/arxiv.2205.09712

arxiv created 2022/05/19 · openalex publication_date 2022/05/19 · arxiv updated 2022/05/20 · openalex created_date 2022/05/23 · openalex updated_date 2026/07/28

Abstract

Large language models (LLMs) have been shown to be capable of impressive few-shot generalisation to new tasks. However, they still tend to perform poorly on multi-step logical reasoning problems. Here we carry out a comprehensive evaluation of LLMs on 50 tasks that probe different aspects of logical reasoning. We show that language models tend to perform fairly well at single step inference or entailment tasks, but struggle to chain together multiple reasoning steps to solve more complex problems. In light of this, we propose a Selection-Inference (SI) framework that exploits pre-trained LLMs as general processing modules, and alternates between selection and inference to generate a series of interpretable, casual reasoning steps leading to the final answer. We show that a 7B parameter LLM used within the SI framework in a 5-shot generalisation setting, with no fine-tuning, yields a performance improvement of over 100% compared to an equivalent vanilla baseline on a suite of 10 logical reasoning tasks. The same model in the same setting even outperforms a significantly larger 280B parameter baseline on the same suite of tasks. Moreover, answers produced by the SI framework are accompanied by a causal natural-language-based reasoning trace, which has important implications for the safety and trustworthiness of the system.

Cited by

• A Large Language Model Based Method for Complex Logical Reasoning over Knowledge Graphs
• On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference
• E-valuator: Reliable Agent Verifiers with Sequential Hypothesis Testing
• Demystifying Errors in LLM Reasoning Traces: An Empirical Study of Code Execution Simulation
• Towards Improving Interpretability of Language Model Generation through a Structured Knowledge Discovery Approach
• INSIGHT: An Interpretable Neural Vision-Language Framework for Reasoning of Generative Artifacts
• Efficient Reasoning via Thought-Training and Thought-Free Inference
• InsurAgent: A Large Language Model-Empowered Agent for Simulating Individual Behavior in Purchasing Flood Insurance
• Video Models Start to Solve Chess, Maze, Sudoku, Mental Rotation, and Raven' Matrices
• Predicate Renaming via Large Language Models
• Credit Cards, Confusion, Computation, and Consequences: What Can We Uncover About Language Model Reasoning?
• Large Language Models Meet Text-Attributed Graphs: A Survey of Integration Frameworks and Applications
• LightMem: Lightweight and Efficient Memory-Augmented Generation
• ActivationReasoning: Logical Reasoning in Latent Activation Spaces
• Toward Mechanistic Explanation of Deductive Reasoning in Language Models
• MEC³O: Multi-Expert Consensus for Code Time Complexity Prediction
• The Idola Tribus of AI: Large Language Models tend to perceive order where none exists
• ARM: Discovering Agentic Reasoning Modules for Generalizable Multi-Agent Systems
• Intra-request branch orchestration for efficient LLM reasoning
• UML-CoT: Structured Reasoning and Planning with Unified Modeling Language for Robotic Room Cleaning
• ReviewScore: Misinformed Peer Review Detection with Large Language Models
• Efficient Graph Understanding with LLMs via Structured Context Injection
• GIER: Gap-Driven Self-Refinement for Large Language Models
• AdaptFlow: Adaptive Workflow Optimization via Meta-Learning
• Simple Methods Defend RAG Systems Well Against Real-World Attacks
• Accelerating LLM Reasoning via Early Rejection with Partial Reward Modeling
• DeepSieve: Information Sieving via LLM-as-a-Knowledge-Router
• Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning
• Transforming Expert Knowledge into Scalable Ontology via Large Language Models
• On the Relevance of Logic for Artificial Intelligence, and the Promise of Neurosymbolic Learning
• Efficient Post-Training Refinement of Latent Reasoning in Large Language Models
• Ctrl-Z Sampling: Diffusion Sampling with Controlled Random Zigzag Explorations
• DipSVD: Dual-importance Protected SVD for Efficient LLM Compression
• From Reproduction to Replication: Evaluating Research Agents with Progressive Code Masking
• OPT-BENCH: Evaluating LLM Agent on Large-Scale Search Spaces Optimization Problems
• Towards Multimodal Graph Large Language Model
• Beyond Accuracy: Dissecting Mathematical Reasoning for LLMs Under Reinforcement Learning
• Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safety
• Dissecting Logical Reasoning in LLMs: A Fine-Grained Evaluation and Supervision Study
• CompileAgent: Automated Real-World Repo-Level Compilation with Tool-Integrated LLM-based Agent System
• Integrating Neural and Symbolic Components in a Model of Pragmatic Question-Answering
• FinRipple: Aligning Large Language Models with Financial Market for Event Ripple Effect Awareness
• ReqBrain: Task-Specific Instruction Tuning of LLMs for AI-Assisted Requirements Generation
• One Model Transfer to All: On Robust Jailbreak Prompts Generation against LLMs
• Teaching Small Language Models to Learn Logic through Meta-Learning
• BAR: A Backward Reasoning based Agent for Complex Minecraft Tasks
• IterKey: Iterative Keyword Generation with LLMs for Enhanced Retrieval Augmented Generation
• Reliable Post-Retrieval Assembly for Agent Memory: Separating Evidence Extraction from Policy Execution
• LogicTree: Structured Proof Exploration for Coherent and Rigorous Logical Reasoning with Large Language Models
• Towards Quantifying Commonsense Reasoning with Mechanistic Insights
• A Survey of Reasoning with Foundation Models: Concepts, Methodologies, and Outlook
• CoLa: Learning to Interactively Collaborate with Large Language Models
• Affordable AI Assistants with Knowledge Graph of Thoughts

Discussions

• Selection-Inference: Exploiting Language Models Interpretable Logical Reasoning [hn, 4 points, 2 comments]

Related

• LogicBench: Towards Systematic Evaluation of Logical Reasoning Ability of Large Language Models
• Are Large Language Models Really Good Logical Reasoners? A Comprehensive Evaluation and Beyond
• Do Large Language Models Excel in Complex Logical Reasoning with Formal Language?
• Logical Reasoning in Large Language Models: A Survey
• Assessing and Enhancing the Robustness of Large Language Models with Task Structure Variations for Logical Reasoning