A stealth-mode startup has emerged with a bold claim: they have cracked the math limiting large language models for nearly a decade. Despite initial skepticism due to sparse details, Subquadratic is backing up their assertions with evidence, potentially reshaping the future of AI efficiency.