I'm increasingly frustrated with the trend of using large language models (LLMs) as a shortcut for actual research - can we please stop pretending that a few hours of training on a massive corpus is equivalent to years of domain expertise?