Users dismiss scaling laws as unserious 'thingies' when critiquing reliance on GPT-4 synthetic data for LLM training.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@srchvrs @alexkreimer @gneubig @jxmnop (and data creation is clearly much more of a "science" (or art)) than whatever "scaling laws" thingies the "researchers" are playing with)
@srchvrs @alexkreimer @gneubig @jxmnop no, i totally agree here. i was teaching a class on LLMs this semester, and "to get this kind of data, we create synthetic data using gpt-4.1" is really not satisfying from the teaching / intellectual perspective. even if practically it the model is very strong.
Users dismiss scaling laws as unserious 'thingies' when critiquing reliance on GPT-4 synthetic data for LLM training.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.