A startup claims it broke through a bottleneck that’s holding back LLMs

Miami-based AI startup Subquadratic has emerged from stealth with a bold claim: it has solved a mathematical bottleneck that has constrained large language models for nearly a decade, according to MIT Tech Review. The company initially provided few details, leading to skepticism, but it has since begun releasing evidence to support its assertion. The bottleneck in question relates to the quadratic scaling of attention mechanisms, which makes LLMs increasingly expensive to run as they grow larger. Subquadratic's approach aims to reduce this complexity, potentially enabling more efficient models without sacrificing performance. For developers and solopreneurs building AI workflows, this breakthrough could lower the cost of deploying and running LLMs, making advanced AI more accessible. However, the claims are still unverified by the broader research community, so caution is warranted. If proven, this could shift the landscape of AI model development, allowing smaller players to compete with larger models from big tech companies.