2 października 2024 r. odbyło się 24 spotkanie LegalTech Polska Meetup. Spotkaliśmy się w warszawskim biurze Bird & Bird na ul. Marszałkowskiej 107, a jednocześnie spotkanie było transmitowane w Teams.
Na spotkanie zarejestrowało się 49 osób, wzięło udział około 30 osób online.
Naszymi gośćmi byli Chris Williams i Tuhin Bhattacharyya z firmy Leya, w związku z czym spotkanie było prowadzone w języku angielskim. Głównym tematem spotkania było przedstawienie nowoczesnego narzędzia, które jest dedykowane prawnikom i które ułatwia korzystanie z systemów generatywnej sztucznej inteligencji.
Żeby otrzymać prezentację ze spotkania, wystarczy wysłać maila na adres tomasz.zalewski@legaltechpolska.pl.
Notatki ze spotkania
Dzięki usłudze Copilot mamy przygotowaną automatycznie notatkę ze spotkania, która została poddana tylko delikatnej edycji.
- Introductions
- Guest Introductions: Chris and Tuhin from Leia were introduced. Chris has spent the last 5-6 years working on innovation in law firms and recently joined Leya to focus on generative AI. Tuhin has an engineering background and experience in AI startups and consulting, having joined Leia when it was a small team.
- Overview of Leia: Leya is a generative AI tool designed for the legal profession. The team discussed its benefits, including its ability to enhance legal work through advanced AI capabilities. Leya is currently being piloted in Bird & Bird and the team aims to share its advantages with the wider legal tech community in Poland.
- Generative AI Limitations: The Leya team discussed the limitations of generative AI models, including hallucinations, prompt engineering, and context windows. They provided examples of hallucinations and explained the reasons behind them.
- Hallucinations: The team explained that hallucinations in generative AI are nonsensical or factually incorrect outputs. They occur due to the statistical nature of language models, which predict the next word in a sequence. Examples included incorrect case details and misinterpretations by AI tools.
- Prompt Engineering: Prompt engineering involves designing inputs to get the best outputs from AI models. It helps control the context and constraints of the models, which is crucial for reducing hallucinations.
- Context Windows: Context windows refer to the amount of text an AI model can process at once. Limitations in context windows can affect the accuracy of outputs, especially for large documents.
- Mitigating Hallucinations: The Leya team explained how they mitigate hallucinations by providing references, accessing specific laws and regulations, and ensuring that the AI does not answer legal research questions without a source.
- Providing References: Leya mitigates hallucinations by always providing references for its outputs. This allows users to verify the sources of information and ensures that the AI’s responses are based on reliable data.
- Accessing Specific Laws: Leya has access to specific laws and regulations depending on the jurisdiction. This ensures that the AI can provide accurate legal information and will not answer legal research questions without a valid source.
- Prompt Engineering: The team discussed the importance of prompt engineering in controlling the context and constraints of generative AI models. They explained how Leia guides users in prompting correctly and how they plan to make prompting more intuitive in the future.
- Importance of Prompt Engineering: Prompt engineering is crucial for controlling the context and constraints of AI models. It helps in reducing hallucinations by ensuring that the AI works with the right data and context.
- Guidance on Prompting: Leya provides guidance to users on how to create effective prompts. This includes prompt libraries and predefined prompts to make the process easier and more efficient.
- Future of Prompting: The team envisions a future where prompting becomes more intuitive and less time-consuming. Users will be able to interact with the AI in a more natural way, similar to communicating with a human.
- Context Windows: The team explained the limitations of context windows in generative AI models and how Leya addresses these limitations by breaking down tasks into manageable pieces.
- Limitations of Context Windows: Context windows limit the amount of text an AI model can process at once. This can affect the accuracy of outputs, especially for large documents. The team provided a visual representation of how context windows have evolved over time.
- Breaking Down Tasks: Leya addresses context window limitations by breaking down tasks into smaller, manageable pieces. This ensures that the AI can maintain accuracy even when working with large documents.
- Demonstration of Leia: The team demonstrated Leya’s capabilities, including summarizing documents, drafting termination letters, and conducting document reviews. They showed how Leya can handle large document sets and provide detailed outputs.
- Summarizing Documents: The demonstration showed how Leya can summarize documents by breaking them down into sections and providing detailed summaries with sentence-level citations. This feature helps users quickly understand the content of large documents.
- Drafting Termination Letters: Leya can draft termination letters by analyzing the context of employment agreements and applying relevant terms. This feature streamlines the process of creating legal documents based on existing agreements.
- Conducting Document Reviews: Leya can conduct document reviews by analyzing multiple documents and extracting relevant information. This feature is particularly useful for tasks like due diligence, where large sets of documents need to be reviewed efficiently.
- Table Review Feature: The team introduced the table review feature, which allows users to review multiple documents and extract specific data points. They explained how this feature can handle large document sets and run reviews in the background.
- Reviewing Multiple Documents: The table review feature allows users to review multiple documents simultaneously and extract specific data points. This feature is designed to handle large document sets efficiently.
- Running Reviews in Background: Reviews can be run in the background, allowing users to continue with other tasks while the review is being conducted. This feature improves productivity by enabling multitasking.
- Comparison Feature: The team demonstrated Leya’s comparison feature, which allows users to compare different contracts and documents. They explained how this feature tailors workflows to specific tasks.
- Comparing Contracts: The comparison feature allows users to compare different contracts and documents. Leya tailors the comparison workflows to the specific tasks at hand, ensuring accurate and relevant comparisons.
- Exporting Outputs: Users can export the outputs of comparisons in various formats, such as Excel or Word documents. This feature provides flexibility in how the comparison results are utilized and shared.