FAQs
Frequently Asked Questions on VerifAI
Q. How does VerifAI utilize LLMs like GPT?
Short Answer:
Our system refines the general language understanding capabilities of LLMs, like GPT, to be more focused on legal contexts. By processing the document and employing mathematical methods, we determine the guideline's relevance to the contract. This allows the LLM to provide responses based on legal knowledge specific to the contract's context.
Detailed Explanation:
- Training Base: LLMs, such as GPT, are trained on vast amounts of public data from the internet. This extensive training enables them to understand language and be aware of publicly accessible information.
- Specialization for Legal Work: While LLMs are knowledgeable, we aim to fine-tune their responses to align more closely with legal knowledge, especially within the context of a specific contract.
- Document Processing: We process the given document, indexing it and transforming its content into embeddings.
- Relevance Determination: Based on the set guideline, we apply mathematical techniques to these embeddings to identify the most pertinent sections of the contract. This step helps ascertain if a guideline is relevant to the contract in question.
- Response Refinement: Once the relevant context is identified, we confine the LLM to this context. We further tune it to cater to legal specifics and ensure it provides answers in a format beneficial to users.
- Action Triggering: The responses generated by the LLM are then used to activate a series of actions and results within our plugin.
Q. What are the underlying models?
Our system is designed to be model-agnostic, ensuring flexibility in adapting to the swiftly evolving landscape of LLMs.
We prioritize harnessing the most recent and robust models to offer the best service to our users. Presently, we predominantly use GPT4 hosted on Microsoft Azure Enterprise for the majority of tasks. However, alternatives such as GPT 3.5, Anthropic, Google PaLM, and Bison are also available.
Q. How is my data used?
- Data Processing: When you engage with a feature in our plugin, it accesses only the document currently open in Microsoft Word and processes it to produce responses. This processing utilizes Google's Firestore Database.
- Data Transmission: During processing, some contextual information from the document is relayed to GPT4. However, as our operations are hosted on Azure Enterprise, Microsoft neither retains nor employs this data for model training.
- Logs & Support: On occasion, we save logs accompanied by timestamps to assist with troubleshooting and offer support in the event of technical anomalies.
- Data Security: Users can rest assured that their credentials are securely stored on Google Firebase.
Q. What are the supported file formats on VerifAI?
Currently, VerifAI doesn't support .doc formats, or documents using compatibility mode.
You can only use .docx files.
Last updated on October 10, 2023