On syntax acquisition in language models, evaluation paradigms, LM representations, and what psycholinguistics and deep learning can do for each other.
I found this conversation fascinating. As an educator, I really like the idea of scaling down models to make closer to human processing capabilities in order to study problem in reading comprehension for instance. Seems like an interesting mode or a vey for educational as well as psychological research.
Agreed! I think there was previously a lot of framing about big/small models as a dichotomy w/r/t/ a singular goal/vision for “the field” and what I like about Prof. Linzen’s perspective is how both are going to be useful, and that we need to carefully sort out what sort of problems we’re studying with which tools.
I found this conversation fascinating. As an educator, I really like the idea of scaling down models to make closer to human processing capabilities in order to study problem in reading comprehension for instance. Seems like an interesting mode or a vey for educational as well as psychological research.
Agreed! I think there was previously a lot of framing about big/small models as a dichotomy w/r/t/ a singular goal/vision for “the field” and what I like about Prof. Linzen’s perspective is how both are going to be useful, and that we need to carefully sort out what sort of problems we’re studying with which tools.