Data security and integrity are top priorities for many businesses and individuals utilizing AI for personalized use-cases. When constructing language models, such as GPT, substantial training datasets are used and potential concerns regarding data privacy may arise. At Ask71, we're committed to addressing these concerns by safeguarding your data and privacy.

In this support article, we will provide insight into the security measures of Ask71 to help you understand how we protect your data at each stage of the process. We'll look into three crucial stages - Documents, Embeddings, and Model.

Ask71 utilizes the secure and private Amazon Simple Storage Service (S3) to store your documents in the initial stage before further processing. S3 ensures encryption of all object uploads to all buckets, maintaining compliance with various programs like PCI-DSS, HIPAA/HITECH, FedRAMP, EU Data Protection Directive, and FISMA. This ensures that your data remains protected and compliant with regulatory requirements. Documents uploaded to Ask71 follow the SSE-S3 (Server-Side Encryption) protocol, allowing exclusive access to you and your team members, ensuring data confidentiality and privacy.

Embeddings are essentially a representation of your data in the form of vectors (lists of numbers). Since the data provided to Ask71 is unstructured, converting it into embeddings allows for faster retrievals and semantic search. To learn more about how Ask71 generates responses from your documents, check out this article.

For storing these vectors or embeddings, Ask71 relies on Pinecone, a secure vector database trusted by some of the largest enterprises.

Pinecone offers robust security features like:

Ask71 leverages OpenAI’s GPT models, including GPT-3.5, GPT-3.5 16K, and GPT-4, to generate responses. Due to resource limitations, these models are not hosted on Ask71’s native servers. Instead they utilise the APIs provided by OpenAI (also used for creating embeddings for your documents and queries). When generating responses, only the specific portion of data relevant to the question asked is sent in the request, rather than transmitting all the documents. This approach ensures efficient processing, data integrity and minimizes unnecessary data transfers. An additional security mechanism provided by the API is that your data will not be used to train any existing or new language model. This ensures that your data remains restricted to your bot and is not utilized for model training purposes.

Source: OpenAI

This commitment provides an additional layer of confidentiality and ensures the privacy and security of your data. To know more, you can read this article.