Abstract: This article first introduces different definitions of “artificial general intelligence” (AGI) in academia and the testing standards for AI generality; then discusses when AGI can be achieved and whether current machine learning technology routes can build AGI; finally, it points out that the main goal of AI development is not LLM-AGI, but “trustworthy and usable AI”. We need more “bounded intelligence”, and a more balanced AI development strategy is a dual-track parallel development model of “data center AGI + specialized AI”.