Comment by firejake308

Counterargument: this works for quick prototyping, but for any serious business, you will eventually develop a benchmark/eval to track how well the general model is working, and once you have that dataset, you might as well train a specific model