The Download: AI bottleneck debates, and BCI trials take off (www.technologyreview.com)

AI startup Subquadratic has emerged from stealth mode announcing a significant breakthrough in addressing a long-standing mathematical bottleneck that has hindered the efficiency of large language models (LLMs) for nearly a decade. By drastically reducing the computation required for transformers to generate responses, Subquadratic claims to deliver a faster, more cost-effective LLM that consumes less energy than its competitors. While many experts remain skeptical of this development, the startup has begun to share data supporting their claims, potentially signaling a transformative moment for the LLM landscape. In parallel, advancements in brain-computer interfaces (BCIs) are making notable strides, with real-life applications beginning to emerge. This week, Casey Harrell, a man living with ALS, became the first person to function as a power user of a new brain implant, which has notably enhanced his ability to communicate and engage with loved ones. The surge in trial participants for BCIs reflects a growing momentum in the sector, highlighted by China's recent approval of a medical-use BCI. The technology's transition from laboratory research to market-ready solutions underscores its potential for revolutionizing patient care and neurotechnology.