On July 20, 2023, a group of 27 scholars and digital rights advocates with
expertise in law, computer science, political science, and other disciplines
gathered for the Large Language Models, Law and Policy Roundtable, co-hosted by
the NYU School of Law's Information Law Institute and the Center for Democracy
& Technology. The roundtable convened to discuss how law and policy can help
address some of the larger societal problems posed by large language models
(LLMs). The discussion focused on three policy topic areas in particular:
1. Truthfulness: What risks do LLMs pose in terms of generating mis- and
disinformation? How can these risks be mitigated from a technical and/or
regulatory perspective?
2. Privacy: What are the biggest privacy risks involved in the creation,
deployment, and use of LLMs? How can these risks be mitigated from a technical
and/or regulatory perspective?
3. Market concentration: What threats do LLMs pose concerning market/power
concentration? How can these risks be mitigated from a technical and/or
regulatory perspective?
In this paper, we provide a detailed summary of the day's proceedings. We
first recap what we deem to be the most important contributions made during the
issue framing discussions. We then provide a list of potential legal and
regulatory interventions generated during the brainstorming discussions.

在这篇论文中，我们详细总结了讨论会的全部过程，包括问题制定的重要贡献以及头脑风暴讨论中提出的潜在法律和监管干预措施。

大型语言模型的监管：圆桌会议报告

Regulating Large Language Models: A Roundtable Report

Mis- and disinformation online have become a major societal problem as major
sources of online harms of different kinds. One common form of mis- and
disinformation is out-of-context (OOC) information, where different pieces of
information are falsely associated, e.g., a real image combined with a false
textual caption or a misleading textual description. Although some past studies
have attempted to defend against OOC mis- and disinformation through external
evidence, they tend to disregard the role of different pieces of evidence with
different stances. Motivated by the intuition that the stance of evidence
represents a bias towards different detection results, we propose a stance
extraction network (SEN) that can extract the stances of different pieces of
multi-modal evidence in a unified framework. Moreover, we introduce a
support-refutation score calculated based on the co-occurrence relations of
named entities into the textual SEN. Extensive experiments on a public
large-scale dataset demonstrated that our proposed method outperformed the
state-of-the-art baselines, with the best model achieving a performance gain of
3.2% in accuracy.

在线误导和虚假信息已成为一个重大社会问题，其中一种常见的形式是错位信息，通过将不同的信息片段错误地关联在一起，例如真实图像与虚假的文字说明或具有误导性的文字描述。本研究提出了一种支持立场提取网络（SEN），可以在一个统一的框架中提取多模态证据的不同立场，并引入基于命名实体的共现关系计算的支持 - 反驳分数到文本 SEN 中。对公共的大规模数据集进行了广泛的实验，结果显示我们的方法优于现有的基线模型，最佳模型在准确率上取得了 3.2% 的性能提升。