In the second quiz, there is a question:
" Question 2
Which of the following are examples of tasks that could be performed by supervised learning models?"
The auto-grader marks as incorrect the item “writing a reply to an email”. But LLM’s can indeed write a reply to an email, and they are trained via supervised learning (next token prediction). Even though the data is self-labeled, it is still fundamentally a supervised learning task is it not?
Training the LMM is supervised learning.
Using an LMM to write an email is not supervised, because there’s no learning required.
You are right LLM can write a reply to an email but the question asks about supervised learning models. LLM are known as Generative models not supervised learning models.
I would argue that “generative models” and “supervised learning models” are not exactly orthogonal concepts, as LLMs are trained via supervised learning, at least in the early phase, but I get now what the question was going for. Thanks!