P2P

62 P E E R T O P E E R : I L T A ' S Q U A R T E R L Y M A G A Z I N E | W I N T E R 2 0 2 3 out of the box. Specifically, the legal industry should not use any model trained by opaque, open-web data full of inaccuracies and biases. By themselves, LLMs cannot validate the accuracy of their output; they can merely infer what they believe to be accurate based on the unvetted information they ingest. This leads to plausible- sounding Gen AI "hallucinations," incorrect or misleading information, and general mistrust of the technology, so much so that judges must ban its use or compel lawyers to attest to the integrity of the citations. Corporate attorneys are increasingly requesting greater transparency and the ability to determine if, when, where, and how AI is used in tools they use to assist in creating their legal work product. Another challenge is setting realistic expectations for what Gen AI can and cannot do (or should not do) and delivering upon them. New technologies that fail to meet the requisite efficiency, productivity, knowledge, or process improvements (beyond what's currently available) or exceed attorneys' exceptionally high standards for privacy, security, and compliance are quickly discarded. According to Gartner's 2023 Hype Cycle for Emerging Technologies, Gen AI has already reached the Peak of Inflated Expectations and is headed towards the Trough of Disillusionment – a period of waning interest as initial experiments and implementations fail to deliver. To reach the Slope of Enlightenment, when a technology's benefits are more clearly understood and materialize, solution providers must rapidly innovate and address the many new challenges that Gen AI creates. We believe growing customer interest and demand and increasing vendor competition will help Gen AI accelerate through this phase. Indeed, Gartner predicts Gen AI will reach transformational benefit (i.e., the Plateau of Productivity) within two to five years – much faster than the traditional seven to ten years for most new technologies. Building Trust with Domain-Specific Responses To overcome these challenges and build trust among users, solution providers need to clearly demonstrate the accuracy and transparency of their Gen AI applications and services, backed by verifiable authority, and minimize instances of invented content. Achieving this requires the creation of legal domain- specific LLMs that prioritize the improvement of model output. This is done using subject matter experts –– attorneys –– to fine-tune models for specific legal use cases; prompt engineering that analyzes a customer's question and adds additional instructions to the model; and integrating vast amounts of caselaw, legal data, news, and other content capabilities using Retrieval-Augmented Generation (RAG) to augment models. RAG extends a model's capacity by connecting it to an external knowledge source. Organizations with access to this high-quality content and pristine data are better positioned to jointly partner F E A T U R E S "Gartner predicts Gen AI will reach transformational benefit (i.e., the Plateau of Productivity) within two to five years."

• Cover