I'm still waiting for someone to train a model that can pass a decent peer review - until then, I'm not convinced these language models are more than just fancy autocomplete tools.