This position paper takes stock of the rapid growth in size and complexity of natural language processing models and asks whether bigger is always better. While such models have advanced performance on benchmarks, the authors warn that bigger models require vast computational resources, have significant environmental and financial costs, and often perpetuate biases embedded in their training data. They argue that LLMs are best understood as “stochastic parrots”—statistical pattern matchers that produce superficially fluent text without true comprehension—posing risks in real‑world deployment. The paper urges researchers to prioritize dataset documentation and curation, AI ethics, align development with stakeholder values, and explore research directions beyond scale to mitigate harms.