Abstract: LLMs are changing how problems are solved, moving from crowd-based platforms like Stack Overflow to AI-driven answers. Our study examines 212,891 problem communities (2020–2025) across four LLM generation model releases (ChatGPT, GPT-4, GPT-4o, o3). Using a difference-in-differences framework, we find that knowledge availability (how much problem-specific information exists) drives the decline in crowd activity. Problem complexity (how interdependent and hard-to-structure a problem is) slows this shift but becomes less protective over time. Results show that communities with more accessible knowledge see steeper drops in activity, while complex problems resist substitution only temporarily. This suggests a temporally moving knowledge frontier, where crowds remain relevant only for problems with scarce knowledge. The shift raises questions about the future of collective knowledge creation in problem-solving domains heavily influenced by AI capabilities.

Bio:Julian Just is a postdoctoral researcher at the Institute of Management and Marketing (Team Innovation and Entrepreneurship) at the University of Innsbruck. His research focuses on the impact of AI on innovation communities, problem-solving, and idea evaluation, with a main focus on quantitative methods such as large-scale text analysis and experiments. He also works for the AI Factory Austria at the Institute of Computer Science, where he fosters the transfer of academic knowledge into practice.

You can read more about him on his website and connect with him on LinkedIn.

Room details: IWI, Room 52-6120 (6th floor)

Online access:

Zoom-Meeting beitreten

https://unisg.zoom.us/j/87224207015?pwd=VDVBYm0yTTFOWWhDMkUvRTdiY2NBUT09

Meeting-ID: 872 2420 7015

Kenncode: 190881

Schnelleinwahl mobil

+13017158592,,87224207015# Vereinigte Staaten von Amerika (Washington

+D.C) 13126266799,,87224207015# Vereinigte Staaten von Amerika (Chicago)