LLMs weren't built for code: one architecture serving prose and strict logic degrades both

gerardsans · x · 2026-10-23

The author argues current LLMs are built around language—open-ended, ambiguous, dynamic—while programming demands strict syntax, grammar, logic and execution flow, two contradictory objectives sharing one unchanged architecture. Asking models to write both Shakespeare and Python without architectural adaptation is why newer models write worse prose; LLMs were designed for English, not code.

Original post →

LLMs weren't built for code: one architecture serving prose and strict logic degrades both

More from Models