Paul Graham Asks Why LLMs Excel at Math But Struggle at Writing: Expert Points to Verifiable Rewards

chris_j_paxton · x · 2026-08-06

Paul Graham tweeted his curiosity about why LLMs have improved dramatically in math and coding but lag in writing. AI researcher Gabriberton quoted the post, explaining that it's not because math is inherently easier. Rather, math and coding have clear right and wrong answers, allowing models to train effectively using verifiable rewards. He noted that experts rarely agree on LLM topics, but the impact of verifiable rewards is a unanimous consensus.

Original post →

More from Models

Models channel →