Unpacking the Power and Challenges of Generative AI in the Legal Industry

A panel of industry experts gathered at the "Generative AI: Uses, Risks and Impacts" session during the Master's Conference in Washington, D.C., to shed light on the confusion surrounding the models driving generative AI within the legal industry. As discussions on leveraging productive artificial intelligence have gained momentum, concerns about the safety of large language models and data privacy have emerged.


During the session, the panelists delved into the fundamental aspects of the large language models powering generative AI while highlighting this technology's associated risks and limitations.

While the term "generative AI" has been frequently discussed among legal professionals in recent months, it has often been used interchangeably and inaccurately with terms like ChatGPT, GPT models, or large language models.

In simple terms, generative AI enables users to input a prompt into a computer program, which then generates text, audio, video, or image outputs, as explained by Aron Ahmadia, senior director for Applied Science at Relativity.

Large language models, a generative AI system, learn from vast amounts of data to predict missing or additional text given some initial input. Ahmadia highlighted that a language model could naturally evolve into a generative model. However, on their own, large language models function more like "all-purpose engines" with inherent limitations.

Tom Shen, team lead of Machine Learning at Bloomberg Industry Group, emphasized the importance of distinguishing between the model and its application when considering large language models. He noted that a layer of code usually directs the language model to perform specific tasks or functions.

Size and data are key factors that differentiate large language models. Shen explained that a model trained on a small amount of data could excel in specific functionalities. However, when it comes to developing versatile tools like ChatGPT, a larger model becomes necessary.

Nevertheless, the legal industry is adopting a "right-sizing" approach to large language models, aiming to distill bigger models into smaller, more cost-effective, and efficient ones. Shen highlighted the need to balance model size and industry requirements.

As the legal industry continues to explore the potential of generative AI, understanding the nuances of large language models and addressing concerns surrounding safety and efficiency will be vital to harnessing the full benefits of this transformative technology.