YANDEXGPT 3 in the hands of a lawyer: what are the benefits for the client?

Neural networks are becoming increasingly popular and are used in many areas of activity. They allow you to speed up work and improve its quality, including in law firms.

According to Thomson and Reuters' Generative AI in Professional Services 2024 report, 35% of respondents from the legal, tax, and government sectors in the US, Canada, UK, Australia, and New Zealand are unsure whether to use generative AI in their work, and 42% of law firms do not plan to use it. However, 25% of organizations have begun to update processes by implementing generative AI, and 47% of respondents are using it themselves for their tasks.

The most common use of generative AI by foreign lawyers is in document management: research, analysis, drafting, and excerpting. According to a 2024 study by the non-profit organization World Commerce & Contracting, lawyers will have even more ways to use AI in the future: they want to use it to assess risks, solve innovative problems, improve negotiation strategies, identify violations of requirements, and analyze documents.

According to the Moscow Digital School 2023 study, 42.9% of lawyers in Russia use ChatGPT in their work. Among them, 43.8% use it to search for the necessary information, 19.2% turn to it to draft standard legal documents, 28% use ChatGPT for marketing and PR, 15.7% - for scientific papers.

In this article, we will look at the example of YandexGPT 3, how a lawyer can use neural networks and what advantages this gives to clients.

Lawyers whose work is closely related to the analysis and processing of texts are increasingly turning to new technologies, such as neural networks, to speed up the fulfillment of client requests. To evaluate the effectiveness of the neural network, two operating modes were used:

• chat mode. In this mode, a lawyer can ask a query limited to 1,000 characters;
• "prompt mode". In this mode, a larger text volume is available - 8,000 characters. Lawyers can manually enter "instructions" ("prompt") to get the answers they need. Each request written in this mode is accompanied by additional options for adjusting the quality and relevance of the received text. For example, you can increase the "temperature" parameter so that the answers are more creative, detailed and original.

Important! It is necessary to strictly observe the confidentiality of information and personal data when working with neural networks, since all requests are recorded. This fact must be taken into account in order to avoid disclosure of sensitive data and the risk of unauthorized access to them.

Various queries (prompts) aimed at accelerating the processing of information for query execution were selected for testing. Here are some examples of tasks that lawyers solved using YandexGPT-3:

• structuring, improving and checking the text;
• translation of text;
• obtaining excerpts and theses from the text;
• selection of options for solving problem situations;
• improving the quality of model responses.

1. Depending on the task, different types of queries were used, which made it possible to obtain the most accurate results.
2. Thus, lawyers have begun to use neural networks in their work to speed up information processing and improve the efficiency of working with clients. However, one must remember about maintaining confidentiality and a critical approach when working with the results.

In legal practice, a specialist often has to work with a large volume of information. Clients expect that a lawyer will be able to quickly and efficiently process the data and offer a solution to the problem.

Is it possible for a lawyer to quickly and efficiently complete a client's task using only general queries? Most likely, no. It is better to turn to ready-made instructions or prompts - templates developed by prompt engineers. They have already proven their effectiveness in practice. Here are some examples of prompts that will help process a client's request faster:

• "Make a summary of the text, provide its main points and the most important facts. [Insert text]";

This instruction allows you to highlight key ideas and important details from any large texts on legal topics. The result obtained can serve as a basis for legal analysis or preparation of analytical reports.

• "edit the text, correcting grammatical and spelling errors, and suggesting ways to improve its readability and clarity. [Insert text]";
• "Act like an experienced editor with many years in the profession. Please provide a list of errors in spelling, punctuation, and grammar, and label these errors. Then, share general thoughts about style and structure. Then, ask for clarification or ask me to offer additional helpful tips. [Text]";
• “Your role is that of an experienced writer and editor. I will give you a text that I have created. I want you to follow a 2-step process. Step 1: I want you to suggest changes to the text that will make it better and more engaging. Think about the structure of the text, sentences, audience, and so on. Feel free to suggest anything that might make the text better. Step 2: I want you to rewrite my text based on the suggestions you gave me in Step 1. Think step by step. Ask any clarifying questions if necessary. My text is: [Text]”;
• "You act as a translator, proofreader, and editor. I will speak to you in any language, and you will identify the language, translate what is said, and produce in response a corrected and improved version of my text. Please translate my text from [language 1], improving it to a more literary version, into [language 2]. Make sure that the version in [language 2] is grammatically and semantically correct. Preserve the original meaning of the text. Write only corrections, improvements, and nothing more, do not write detailed explanations."

The above prompts are useful and effective for reducing the lawyer's time spent on proofreading texts and improving their structure. As a result, the client receives competently composed documents written in understandable language.

• "critically analyze my idea. Point out three pain points of the idea. But don't just point them out, immediately give one or two options for solving the problem. Idea: [text]";
• "ask 5 tricky questions about my idea so that I can refine it. Answer these questions yourself, offer two possible solutions for each point: a standard one and a maximally unusual one";
• Describe the process in detail."

These prompts help the lawyer speed up the process of finding solutions and reduce the time it takes to develop the optimal plan of action for the client.

Important! When drafting documents or texts containing legal norms, you need to be careful. In such cases, YandexGPT 3 can provide either general phrases that do not carry specific legal information, or fictitious links to non-existent sources of law. It is recommended that a lawyer use the data obtained from the neural network as a basis for the future text. All legal norms and current examples from judicial practice must be checked and added independently.

The main conclusion: neural networks in the legal sphere will find their niche. Now they are especially useful when working with large volumes of text, as they allow the client to receive a high-quality answer in a short time. The text generated by the neural network requires the attention and verification of a lawyer, as the neural network does not take into account the context of the request and is not trained to answer legal questions. To improve the quality of YandexGPT 3, you can use prompts, but they will not make the neural network work flawlessly. The main advantage of using neural networks is the reduction in the time that a lawyer spends on reading and analyzing large texts, as well as on forming ideas and solutions to the client's requests. The neural network makes informative excerpts, processes the text, identifies errors, offers ideas for solving problematic situations, speeding up the specialist's work and reducing the client's costs for legal services.

However, you should not rely entirely on artificial intelligence in solving issues and drafting documents, as it does not take into account the context and circumstances that are important for resolving problematic situations. The final decision and content of the text is determined by a person.

Interestingly, SBER is developing GigaLegal, a neural network for lawyers that can already analyze short text for risks and answer questions from the knowledge base. In the near future, GigaLegal will allow extracting information, depersonalizing documents, generating documents, identifying legal risks, briefly retelling documents and answering queries about them. Doczilla supplements the artificial intelligence module with an assistant for branches of law and is testing it in a closed mode of the tax legislation reference book.

Thus, specialized neural networks for lawyers are still in the learning process, but we can already look with optimism to the future of using artificial intelligence by lawyers to effectively build work with clients.