Researchers · Interviewed at COLM 2026
Carnegie Mellon University / DatologyAI

COLM 2026 · October 7, 2026 · 23 min
Most teams with a small domain dataset follow the standard recipe: pretrain a general model, then fine-tune on the domain data. Christina Baek's results suggest that recipe leaves performance on the table.
Work is presented at COLM 2026 (the Third Annual Conference on Language Modeling).