The Download: AI bottleneck debates, and BCI trials take off
AI startup Subquadratic announced a breakthrough in solving a mathematical bottleneck that has limited large language models (LLMs) for nearly a decade. The company claims its new approach significantly reduces the number of computations required for transformers to generate responses, leading to faster, cheaper, and more energy-efficient LLMs. While many experts remain skeptical, Subquadratic has begun sharing data to support its claims, suggesting its method warrants attention. The technology aims to improve the efficiency of LLMs by reducing computational load.
Brain-computer interface (BCI) trials are experiencing a surge in activity. Casey Harrell, a man with ALS, is highlighted as the "first power user" of a brain implant that has allowed him to work, communicate with loved ones, and read to his daughter, describing the experience as "nothing short of revolutionary." The number of BCI trial volunteers has increased significantly in recent years. In 2024, China became the first nation to approve a BCI for medical applications, signaling a move of the technology from research labs toward broader market adoption.
In other technology news, Amazon workers who supported limits on data center expansion may face termination, with engineers reportedly under company investigation. This development raises concerns about the company's stance on employee activism regarding environmental and infrastructure issues.
Original source — read the full reporting at the publisher:
Read on MIT Technology Review