Subquadratic claims to solve decade-old LLM math bottleneck
subquadratic slashed transformer computations to create faster, cheaper llms, but experts remain skeptical.
subquadratic emerged from stealth with a claim it solved a mathematical bottleneck limiting llms by slashing transformer computations. the startup asserts the resulting model is faster, cheaper, and more energy-efficient than existing options. while some researchers have seen data suggesting the approach works, many experts remain skeptical. if true, the breakthrough could drastically lower the costs of running generative ai.