TapTechNews on June 29th, via the official WeChat public accounts of Mianbi Intelligence and Shenzhen Intermediate People's Court, the first large-scale model in the judicial trial vertical field in China assisted by Mianbi Intelligence Company was born. The AI-assisted trial system went online and started operating in the Shenzhen Intermediate People's Court on the 28th.
According to the introduction, this large-scale model can accurately diagnose case situations and solve the application difficulties of AI in the judicial field. This system can fully respect the autonomous decision-making power of judges and ensure that technological progress does not overstep its authority, so that judicial judgments are always made by judicial personnel.
This system can comprehensively cover 85 processes of trial business such as case-filing, case-reading, court hearings, and document production and realize AI's full-chain empowerment.
At each link node of this system, there are audit, confirmation, decision options, and prompts set up, and these serve as preconditions for AI-assisted generation. AI prompts judges to make judgments on doubtful points according to the intelligent comparison and investigation situation, and assists in generating judgment documents according to the judgment result decided by the judge. The process can be corrected in a timely manner and can be traced back throughout the whole process.
This system has eight highlights including full-process AI empowerment, material entry factorization, initiative tree-structured prompt words project, authoritative knowledge service system, standardized judicial thinking link, precise data of source governance of litigation, compatibility and expansion function of technical advancedness. Since the trial operation of the system in January this year, it has assisted in filing 291,000 cases; and assisted in generating 11,600 draft documents.
According to TapTechNews' previous report, Mianbi's core R&D team is derived from the Tsinghua Natural Language Processing Laboratory. Previously, their open-source model Little Cannon MiniCPM-Llama3-V2.5 was plagiarized by the Llama3-V open-source model of Stanford University's AI research team.