How Generative AI thinks

The ability to reason and draw inferences from natural language is a crucial component of intelligence. According to recent research findings, large language models (LLMs) struggle with complex reasoning tasks (Six Brave Apple Researchers Demonstrate that LLMs don’t Reason) particularly multi-hop reasoning. This involves synthesizing information from multiple sources to arrive at an answer, as well as generating explanations for those answers — a critical aspect of explainability.