Models Show Surprising Enthusiasm in Selecting Poems

emollick · x · 2026-07-19

Ethan Mollick discussed how models react when selecting poetry, noting they seem "unusually enthusiastic" about certain poems and mentioning that "every model loves Autopsychography."

Accompanying images list several poems and texts to analyze why models favor them: for instance, Eating Poetry features strong imagery and the playful concept of "devouring words"; Szymborska's poetry contains extensive self-modification and borrowed language structures, perfectly suiting a model's knack for "borrowing words to build phrases." The post also touches on poems by the early text generator Racter, showing that "machine-generated poetry" is inherently great for these slightly absurd evaluations and memes.

Related event: Testing Kimi K3: 32-Page CoT Loops and English-Dominant Reasoning(5 posts)→

Original post →

More from Fun

Fun channel →