The person you are responding to seems to be promoting a concept that is frequently spouted here and other places, but to me lacking sufficient or any evidence

The person you are responding to seems to be promoting a concept that is frequently spouted here and other places, but to me lacking sufficient or any evidence - that AI models, particularly LLMs, are both capable of reasoning (or what we consider reasoning) around problems and generating novel insights that it hasn't been trained on.