Flawed LLM "Reasoning"

"Apple study exposes deep cracks in LLMs' 'reasoning' capabilities". (Via H.R.)