These LLMs are confabulation machines. They're as good as the knowledge of the person who is driving the current session. For coding problems this makes them like a very good teddy bear debugger, but do not expect independently produced creative work from them.