速递｜OpenAI前研究副总裁自立门户：新实验室筹集5至10亿美元融资

Core Insights - The article discusses the ongoing trend of emerging AI labs, particularly focusing on Core Automation, founded by former OpenAI researcher Jerry Tworek, which aims to raise $500 million to $1 billion for developing AI models that can learn continuously from real-world experiences, a capability current models lack [1][2] - Tworek represents a growing group of AI researchers advocating for a complete overhaul of existing model development techniques, as they believe current popular methods are unlikely to yield advanced AI breakthroughs in fields like biology and pharmaceuticals while avoiding basic errors [2][3] - Core Automation plans to utilize large neural networks but intends to rethink the development process, including the standard training methods like gradient descent, aiming to create models that require significantly less data and server resources for training [2][3] Company Development Plans - Tworek envisions a model named Ceres, developed through a single algorithm, contrasting with the phased training approach typically used by large AI developers, which involves pre-training on vast internet data followed by specialized training [4][6] - The company aims to create an AI agent to automate product development, with initial applications in industrial automation, ultimately aspiring to build "self-replicating factories" capable of producing biological machines for custom designs [6] Industry Context - The interest in continuous learning technology is shared by other AI labs, such as the Safety Superintelligence Lab co-founded by former OpenAI chief scientist Ilya Sutskever, indicating a broader industry trend [3] - Despite the lack of revenue or products from many of these emerging labs, investor interest remains strong, as evidenced by recent funding rounds for companies like Humans& and Thinking Machines Lab, which have raised substantial amounts [3]