> Tuned models are expensive to produce The expensive part is building the dataset, training itself isn't too expensive (you can even fine-tune small models on

> Tuned models are expensive to produce The expensive part is building the dataset, training itself isn't too expensive (you can even fine-tune small models on free colab instances!), and when you have your dataset, you can just fine tune the next generalist model as soon as it's released and you're good to go ago.