AI Startup Subquadratic Claims Breakthrough in LLM Efficiency, Solving Decade-Old Bottleneck
A startup claims to have cracked a decade-old AI bottleneck: Miami-based Subquadratic says its new model, SubQ, solves the computational problem that makes LLMs slow and expensive. Many people are skeptical, but the company has now brought in outside validators to back up its bold claims. The math behind the breakthrough: Most LLMs multiply every word against every other word in a document, causing costs to explode as text gets longer. SubQ uses "sparse attention" to skip unn
Miami-based startup Subquadratic asserts it has overcome a significant computational bottleneck that has long plagued large language models (LLMs), making them slow and costly. Their new model, SubQ, reportedly utilizes "sparse attention" to bypass unnecessary word comparisons, a technique that has eluded previous efforts. While initial claims were met with skepticism, an independent evaluation by Appen found SubQ to be 56 times faster than rivals and achieve 98% accuracy on a key long-document retrieval test. The company claims SubQ can process 12 times more text simultaneously than existing models, enabling complex data-heavy tasks. However, the model is not yet widely available, and its development on borrowed weights from an open-source Chinese model has raised questions.
If validated, Subquadratic's breakthrough could dramatically reduce the cost and increase the speed of large language models, accelerating AI development and adoption across various industries.
📌 Kaynak
Bu haber XML kaynağından derlenmiştir. Tamamı için orijinal habere gidin.
Orijinal haberi oku →