China's ByteDance discovers new scaling law that could sustain AI boom (www.scmp.com)

ByteDance's Seed AI team has made a significant breakthrough in autonomous AI development by identifying a new scaling law that potentially accelerates the learning process of AI agents. Their research, published recently, indicates that these AI systems can double their learning speed every three months when they engage with real-world environments over extended periods. This discovery is particularly crucial as the AI industry seeks innovative methods for enhancing model performance, especially amidst growing concerns that traditional approaches—relying heavily on increased data and computing power—may not be sustainable long-term. To further explore this concept, the researchers introduced EdgeBench, a comprehensive benchmarking suite consisting of 134 ultra-long-horizon tasks across diverse fields such as software engineering and scientific discovery. Each task is designed for a minimum of 12 hours of continuous operation, providing a robust framework to examine how AI agents can learn and adapt post-deployment. This advancement not only offers a clearer understanding of agentic AI but also sets the stage for more effective and efficient AI systems capable of handling complex real-world applications, thereby driving the next wave of innovation in the field.