Expert Roundtable on the Impact of GPT‑4 and Large Models on Knowledge Graphs

The interview selects key topics from a Knowledge Graph summit roundtable, gathering hard‑core viewpoints from domain experts on how GPT‑4 and large models influence knowledge graphs and AI development.

Q1 explores the impact of GPT‑4’s multimodal capabilities and its ability to solve complex, human‑level tasks such as lawyer and Olympiad exams, highlighting longer context windows and improved reasoning.

Q2 discusses the trend toward unified multimodal AI that integrates image, text, speech, and NLP, emphasizing that future models will increasingly resemble human perception.

Q3 argues that knowledge graphs remain essential for factual accuracy, verification, and domain‑specific knowledge that large models alone cannot reliably capture.

Q4 examines how knowledge graphs can aid large models in fact‑checking and providing up‑to‑date, verifiable information through external retrieval mechanisms such as vector databases.

Q5 outlines practical ways to incorporate knowledge graphs into large‑model pipelines, including converting graph data to text for training, preserving structured information for retrieval, and using external tools or plug‑ins during inference.

Q6 considers whether traditional NLP tasks (entity recognition, relation extraction, etc.) are still needed, concluding they remain valuable for domain adaptation and fine‑grained analysis despite end‑to‑end model advances.

Q7 provides an overview of China’s large‑model landscape, noting industry efforts at model scaling, algorithmic innovation, and the integration of knowledge graphs within commercial AI products.

Q8 offers concise advice for practitioners: embrace large‑model advances while tightly coupling them with domain expertise and knowledge‑graph resources to achieve optimal results.

The article concludes with brief biographies of the participating experts—professors and senior AI leaders from Harbin Institute of Technology, Tongji University, iFlytek, and Daguan Data—highlighting their research focus on knowledge graphs, natural language processing, and AI engineering.