What will happen as more and more LLM-generated text gets swept up into LLM training sets? This workshop examines the nature of LLMs and authoritative text.

As part of the Hesburgh Libraries AI Literacy workshop series, this installment addresses the practical concern that self-reference of AI-generated materials in AI training sets may cause Large Language Models (LLMs) to deteriorate over time.

 Where did that idea come from? How much are LLMs muddying the water regarding the provenance of ideas? How much are they increasing the spread of misinformation and disinformation?

What will happen as more and more LLM-generated text gets swept up into LLM training sets? If the purpose of LLMs is to replicate human use of language, can LLMs continue to improve if they can’t screen AI-generated material out of subsequent iterations of upgraded software?

How should one fact-check AI? If AI provides no references, what can be done to determine the authoritative nature of an AI-generated text? If you find these questions intriguing or troublesome, attend this workshop and examine the nature of LLMs and authoritative text.

The AI Literacy workshop series is an offering of the AI Collaborative Programming group, a cross-unit partnership between the Hesburgh Libraries, ND Learning, and the Office of Information Technology.