@patrickqdasilva
Student Researcher @Google PhD Student at @OhioStateCSE
AI
The author says language-model editors often made strong abstracts worse and pointed readers to a COLM 2026 poster.