Xuan Gong’s research while affiliated with Zhejiang University and other places

Pre-trained large language models (LLMs), epitomized by ChatGPT, have leveraged a cornucopia of “big data” to attain substantial leaps in artificial intelligence (AI). Whereas the diminishing returns from pre-training and the depletion of available training data have become evident, the post-training scaling law bolstered by “big GPU” has surfaced as an overriding strategy. Since 2024, post-trained models exemplified by o1 and DeepSeek-R1 have been widely acclaimed as successes in logic-intensive fields like advanced scientific problem-solving, serving as a bellwether for artificial general intelligence (AGI). Driven by the two cardinal elements of computing power and task-specific datasets, the data training processes of post-trained models exhibit more erratic and uncontrollable tendencies, which may be a menace to core societal domains and precipitate systemic friction vis-à-vis the existing data governance derived from pre-trained models. At this watershed moment, this article aims to conduct a comprehensive comparison of training data paradigms between pre-trained and post-trained models and to further develop cogent and favorable governance responses to mitigate emerging risks. Consequently, data security must be established as a prerequisite for AI development, and a lifecycle-based governance framework for AI training data in blended models can be introduced in the metamorphosis toward “bigger AI models”.

The explosive advancement of contemporary artificial intelligence (AI) technologies, typified by ChatGPT, is steering humans towards an uncontrollable trajectory to artificial general intelligence (AGI). Against the backdrop of a series of transformative breakthroughs, big tech companies such as OpenAI and Google have initiated an “AGI race” on a supranational level. As technological power becomes increasingly absolute, structural challenges may erupt with an unprecedented velocity, potentially resulting in disorderly expansion and even malignant development of AI technologies. To preserve the dignity and safety of human-beings in a brand-new AGI epoch, it is imperative to implement regulatory guidelines to limit the applications of AGI within the confines of human ethics and rules to further counteract the potential downsides. To promote the benevolent evolution of AGI, the principles of Humanism should be underscored and the connotation of Digital Humanism should be further enriched. Correspondingly, the current regulatory paradigm for generative AI may also be overhauled under the tenet of Digital Humanism to adapt to the quantum leaps and subversive shifts produced by AGI in the future. Positioned at the nexus of legal studies, computer science, and moral philosophy, this study therefore charts a course for a synthetic regulation framework of AGI under Digital Humanism.