Large Language Model Reasoning Failures

2026

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

1) This survey provides a comprehensive overview of the various ways Large Language Models (LLMs) fail at reasoning, offering a structured framework for understanding and addressing these critical limitations.
2)
* Categorizes LLM reasoning failures into embodied and non-embodied types, further subdividing non-embodied reasoning into informal and formal.
* Classifies failures based on their nature: fundamental, application-specific, and robustness issues.
* Analyzes existing research, identifies root causes, and suggests mitigation strategies for each failure type.
3) LLM reasoning, AI failures, language models, reasoning taxonomy, AI safety

Check the original publication for accuracy and context.