[winddude]

It's this one "these LLM-categories are a bit dubious", specialized models still outperform LLMs on niche tasks like classification and sentiment.