Large language models,represented by ChatGPT and ERNIE Bot,have generated a substantial volume of large language model materials,which have permeated society and had a profound impact.Therefore,dis-cussing the evidentiary attributes of these materials is of significant importance.Large language model materi-als refer to materials related to the usage of large language models that may be used as evidence in legal pro-ceedings.Narrowly defined,large language model materials exclusively encompass materials generated by large language models.In a broader sense,they include human-machine interaction materials,the large lan-guage model itself,and information about the operational context of the large language model,comprising these three components.These materials can be categorized into different types based on the degree to which they reflect"human-machine opinions"and their role in the evidentiary process.According to the definition of evidence in Chinese law,large language model materials that can establish the facts of a case qualify as evidence.To determine the admissibility of large language model evidence,its ob-jectivity,relevance,and legality must be examined.Large language model evidence reflects objective facts in its content,is perceptibly objective in its form,demonstrating objectivity.It is deeply integrated into social life,connecting with the facts of cases in various scenarios,such as civil,criminal,and administrative cases,exhibiting relevance.While it currently lacks a legal foundation,it can acquire legality through the enhance-ment of the legal framework.Therefore,the question of whether large language model evidence qualifies as evidence requires specific consideration in future cases.For the sake of litigation efficiency,the presentation of large language model evidence may vary in different litigation scenarios.In standard cases,only human-ma-chine interaction materials and the user's local context information may suffice,while significant cases may require a comprehensive presentation of large language model evidence.With the advancement of cutting-edge technology,new types of technological evidence have emerged,in-cluding big data evidence and artificial intelligence evidence.Big data evidence and large language model evi-dence face similar challenges related to the"black box",admissibility,and categorization.Additionally,big data evidence has been widely utilized in practice,making the associated theories a valuable reference for the development of large language model evidence.While artificial intelligence evidence is a higher-level concept than large language model evidence,the extensive diversity in the field of artificial intelligence technology im-plies that the findings of AI evidence research may not necessarily be directly applicable to large language model evidence.Large language model evidence exhibits the following characteristics.Firstly,it possesses a high degree of intuitiveness,as materials pertaining to human-machine interaction in large language model evidence can be readily perceived by humans.Secondly,it has limited interpretability due to inherent black-box effects and e-mergent characteristics,resulting in its relatively weak interpretability.Thirdly,there may be evidence bias in large language model evidence,potentially favoring a select few technology companies.Our legal system can draw inspiration from the electronic data disclosure systems in the Anglo-American legal tradition to establish disclosure obligations for technology companies.Finally,it has limited identifiability,as humans may not nec-essarily be able to discern whether a piece of material was generated by a large language model in the absence of specific indicators.This issue can potentially be mitigated through the deep synthetic regulation.
NETL
NSTL
万方数据
