A Modular Dataset to Demonstrate LLM Abstraction Capability

Adam Atanas, Kai Liu

Large language models (LLMs) exhibit impressive capabilities but struggle with reasoning errors due to hallucinations and flawed logic. To investigate their internal representation…