Opening Remarks

The Sources of LLM Data

The first panel of the conference will examine how designers and developers acquire the datasets used to train LLMs and explore whether there are alternative strategies to obtain this data that can yield more complete, accurate, and representative datasets. Panelists will also explore the impact of web scraping on individuals and groups and will assess whether content creators should be granted new protections for the reuse of their online content.

Keynote: The Potential of Synthetic Data for Generative AI

The Continuum of Closed and Open LLMs and their Implications for Data Governance

This panel will examine the implications of open and closed LLMs for effective data governance. We will discuss how LLM designers, developers, and deployers make decisions about openness and in so doing, navigate the spectrum from closed systems, where algorithms and data remain tightly held, to varying degrees of openness. Open-source models offer greater transparency, which can provide outsiders with insights into data provenance, validity, completeness, and representativeness. Open datasets allow content creators to gain insight into whether their works were used to train the model. However, open datasets and open LLMs come with trade-offs in terms of LLM stability and safety.

Data Openness and Society

This panel will examine the long-term societal implications of the choices data openness, exploring their implications for open science, trust, democracy, and the future course of AI development. While open data is essential to democratic accountability and scientific replicability of research, some advocates of openness are increasingly concerned about its unintended side effects, including misrepresentation or lies about scientific data or historical facts.

Day 1 Wrap-Up and Consensus-Building

Opening Remarks

What Are Governments Doing to Address the Data Governance Challenges Presented by LLMs?

Fireside Chat: Data as a Civics Issue

New Ideas for Shared Data Governance

The final discussion of the conference will build on the earlier discussions of how to address the enforcement problems and governance gaps revealed by generative AI. We will also discuss how to promote greater understanding of, and participation in, the governance of data.

Conference Wrap-Up Discussion and Consensus-Building