AI systems often produce unreliable information, a concern that particularly affects businesses where inaccuracies can impact profitability. A recent Salesforce survey found that half of employees worry about the accuracy of responses generated by their company’s AI systems. While there is no foolproof method to eliminate these “hallucinations,” certain techniques can mitigate them. One such technique is retrieval-augmented generation (RAG), which combines an AI model with a knowledge base to provide supplementary information before generating answers, effectively acting as a fact-checking mechanism.
Voyage AI, founded in 2023 by Stanford professor Tengyu Ma, is one of the companies leveraging RAG technology to enhance AI reliability. The company provides RAG solutions for various clients, including Harvey, Vanta, Replit, and SK Telecom. Ma describes Voyage’s mission as improving search and retrieval accuracy in enterprise AI, with solutions tailored to specific domains such as coding, finance, legal matters, and multilingual applications5.
How Voyage AI Works
Voyage AI develops RAG systems by training models to transform various data typesālike text documents and PDFsāinto numerical representations known asĀ vector embeddings. These embeddings capture the meanings and relationships between data points in a compact form, which is crucial for search applications. A unique aspect of Voyage’s approach is its use ofĀ contextual embeddings, which not only understand the semantic meaning of words but also their context. For instance, the word “bank” would be represented differently depending on whether it’s used in a financial context or a geographical one.Voyage offers its models for deployment in various environments, including on-premises and cloud solutions. The company also provides fine-tuning services for clients seeking customized models. While other companies like OpenAI offer similar services, Ma asserts that Voyage’s models achieve superior performance at lower costs.
Performance and Recognition
Ma highlights that conventional RAG methods often encounter issues with context loss during information encoding, which can hinder the retrieval of relevant information. However, Voyage’s embedding models are designed to excel in retrieval accuracy, significantly enhancing the overall quality of responses generated by RAG systems. This claim is supported by an endorsement fromĀ Anthropic, a competitor of OpenAI, which describes Voyage’s models as āstate of the artā.As of now, Voyage has over 250 customers and recently completed a $20 million Series A funding round led by CRV, with contributions from investors like Wing VC and Snowflake. This funding will facilitate the development of new embedding models and allow the company to expand its workforce.
Spread the loveMicrosoft continues to offer a diverse selection of courses for beginners in 2024, aimed at individuals eager to improve their skills in applications like Word, Excel, PowerPoint, and Access. These courses are structured Read more…
Spread the loveLawctopus, a premier platform for law students and young professionals in India, is thrilled to announce aĀ Video Editing Internship. This remote opportunity allows creative individuals passionate about visual storytelling to contribute to the Read more…
Spread the loveTata Consultancy Services (TCS) is offering a diverse array ofĀ free online coursesĀ in 2024, specifically tailored for fresh graduates looking to enhance their skills and boost their employability. Recognizing the rapid changes in technology Read more…
0 Comments