2024-02-20 20:00:00
Called by the general directorate of public finances, around thirty French researchers have developed artificial intelligence capable of synthesizing a ton of legal texts. To do this, they relied on Meta’s Llama 2 model, and the first results are there.
Between the very specific jargon of the law and the phenomenal quantity of amendments that can be proposed to a text of law, it is not an easy task to understand what is decided in the Assembly and the Senate.
Rather than reducing or simplifying the content itself, the Directorate General of Public Finances (DGFiP) has chosen to use artificial intelligence (AI) to fully analyze and summarize it. With the help of the Legal and Administrative Information Directorate (Dila) and the Interministerial Digital Directorate (Dinum), she designed LlaMandement.
The tool ingested some 15,000 amendments, all written in the same way, thus becoming a machine responding “the administrative challenges posed by the manual processing of an increasing volume of legislative amendments […] while matching the robustness of a specialized legal editor”. It identifies the parts, summarizes the purpose of the text and compares it to the others.
As reported by Siècle Digital, LlaMandement has already been tested for the study of the 2024 Finance bill. Among all the amendments tabled in 2023, it quickly recognized those which were linked to this bill , even specifying that one text in six was redundant. Then he summarized them, almost as well as the specialist agents. It is therefore possible that this tool will be more widely used in the future.
#summarize #legal #texts