P2P

78 By contrast, white-box AI is designed for transparency, allowing users insights into the logic and decision- making process. This openness helps users see how the AI arrives at its decisions. The transparency of white-box AI enables developers, users, and regulators to examine, validate, and adjust the AI's behavior to ensure accuracy, fairness, and ethical compliance. This clarity builds trust and accountability, empowering users to better understand and manage the system, much like having a detailed instruction manual. In the legal context, where inaccuracies present a significant risk to law firms, it is obvious that white-box AI solutions are preferred. Think of it this way: before AI, citation checks were a normal and essential part of the legal writing process. It is the same now. Legal AI solutions can accelerate information gathering, making the drafting process easier; however, they should not replace human oversight altogether. While some GenAI solutions are designed to cross- reference outputs with authoritative legal sources, such as case law databases, statutes, regulations, and secondary sources, effective prompt engineering can help ensure that outputs are connected to case data or other verifiable sources and are not hallucinations that are just strings of predicted texts not rooted in reality. For example, you can ask AI to analyze a set of deposition transcripts and identify inconsistencies in several witnesses' testimony. Even better, tell AI to do the same thing and link back to relevant passages in the testimony. That output will provide citations that can be easily verified to ensure the GenAI analysis is accurate and reliable. With a basic understanding of how LLMs and the prompt-output process work, you can start creating prompts to help automate your work. STRATEGIES FOR BUILDING BETTER PROMPTS As we said before, prompt engineering works best as an iterative process. As you prompt and re-prompt your AI model, think about employing one or more of these strategies as you tweak your prompts. Begin by informing the AI about yourself and your intended purpose, including the target audience for the output you wish to generate. For example, you could say, "I am a lawyer preparing for a deposition." But an even better prompt would be: "I am an intellectual property attorney preparing for a deposition with an inventor of semiconductor technology." This sets the stage and helps the AI understand the kind of output you're most likely looking for. Do not write like a lawyer: Lawyers are great at writing like lawyers, but writing to AI in plain language will result in a better response. Why? The foundational training for most AI is based on day-to-day writing and speaking samples, not legalese. Social media illustrates this idea in "Explain it to me like I am five" posts, in which experts in their field offer plain-language insights on complex topics, so that a layperson can understand. When working with AI, keep this idea in mind to help you reframe legal concepts when constructing your prompts. Give feedback: If the generated response is not what you were looking for, do not give up. Tell the AI what parts you liked and why you didn't like the other parts. If the response does not include citations, ask for them. Use a conversational tone: If the AI generates a puzzling response, again, do not give up. Ask your question in a different way, just like you would to a friend or a colleague who said something you found confusing.

• Cover