Topical Phase Transitions in Artificial Intelligence Research: Large-Scale Evidence and an Early-Warning Signature for Emerging Topics

In three linesAnalysis of 80,814 papers from 5 major AI conferences (2017-2025) reveals research topics advance through abrupt phase transitions, not gradually. LLMs dominant by 2025; diffusion models and vision-language models surged within 1-3 years. Early-warning signature flags reasoning, test-time compute, agentic AI, multimodal LLMs, RAG, and world models as topics to monitor 2026-2028.

Summary generated by Claude — human-verified