The phenomena that allows for Reasoning in modern LLM when they are scaled up. Older LLMs could only translate Adding more attention layers allows it to Reason