Theory: Typos Cause Slight Discomfort to LLMs by Disrupting Tokenization
cosine_distance · x · 2026-11-11
cosinedistance proposes that typos disrupt LLM tokenization and cause the models slight "discomfort," offering a lens on why AI text tends to be linguistically over-clean compared to human writing.
Related event: Do typos annoy LLMs? Tokenization theory sparks debate(5 posts)→