Can AI Replace Humans? — How Human Knowledge Continues to Grow in the Era of Large Language Models

Abstract

Recommendation algorithms determine what people see, and may also influence the perspectives through which people understand the world. As Large Language Models (LLMs) enter the domains of knowledge acquisition and information comprehension, this influence may extend further into human understanding, judgment, and modes of thinking. If people rely long-term on a small number of general-purpose LLMs, a new homogenization of knowledge sources and modes of interpretation may emerge. A possible direction is the joint development of general-purpose LLMs and vertical small models: general models provide breadth, while vertical models leverage industry-specific and professional data to provide depth and novelty. LLMs as knowledge tools do not imply that humans will be replaced. LLMs provide the knowledge foundation and computational power; humans provide direction. In this collaborative process, individuals internalize knowledge and continuously push toward the unknown through judgment, reasoning, verification, reorganization, and Global Self-Consistency. When a problem reaches the point where existing knowledge can no longer explain it, new thinking emerges naturally. As more and more individuals explore in different directions, innovation points accumulate and connect, ultimately forming a new knowledge foundation. When there are enough points, they connect into surfaces; when there are enough surfaces, they form a new knowledge foundation.