Those are two different things. GPT-3 is a lang model. so it can generate sentences which looks like real human.
laMDA was trained on dialogs, so it can converse. I.e. it more attune to conversation which has a different structure than just question-answering or text generation.