Generative AI and Authorship

Generative AI = Large Language Models

Large Language Models (LLMs) = Data

According to Prof. Vivek Srikumar LLM Data = 

• Entire Internet
• All Books from last 100 years

As a data set, generative AI has the following limitations and conditions: 

• can’t assess and prioritize sources

• can’t build knowledge

• can’t make meaning

• output is sensitive to starting conditions

AI companies have built up around LLMs. Each company will have their own terms of use and publication policies:

• OpenAI: "Manually review each generation before sharing or while streaming; Attribute the content to your name or your company."
• Hugging Face (no publication policy)
• StabilityAI (no publication policy)
• LumaAI: "You agree that you must evaluate, and bear all risks associated with, the use of any content, including any reliance on the accuracy, completeness, or usefulness of such content."

When it comes to scholarly communication and publishing, there will be expectations about acknowledging authorship; see Publisher Policies page for a sampling.