By my expirence and by opinion a lot of people in NLP, with which I had conversation, prompt learning to llm from the box can be enough for at least 50% of tasks.
You have just one place where you hint on it — Hugging Face labs. It would be nice to say more obvious about this in NLP courses. I think it would be pertinentely to add courses like this or this to your NLP courses.
You’re right. However, understanding what happens behind the scenes allows for better performance optimization, issue diagnosis, and the most important of all: innovation.
I agree. But I didn’t feel myself ready for real NLP problems solutions after the course — this is not ok. Your approaches are exessive for a lot of tasks.
Also, would be good to mention what data structure to use for different tasks, if I am free to choose it when collect the data.
Also, I think the big space in your course — that you don’t introduce us with any label systems. I think, we should start each of courses (that are connected to specific type of tasks) from it. For example — Label Studio is good one to mention. They have good interface at least for classification and ner tasks.