Reasoning is present in language models, but it's difficult to determine whether the model is performing emulated reasoning based on human-written reasoning in training data or developing reasoning independently, and since humans also learn reasoning by reading about it, this distinction may be philosophically unclear.