AI tends to make issues up. That’s unappealing to only about anybody who makes use of it frequently, however particularly to companies, for which fallacious outcomes may damage the underside line. Half of staff responding to a current survey from Salesforce say they fear solutions from their firm’s generative AI-powered techniques are inaccurate.
Whereas no method can clear up these “hallucinations,” some may also help. For instance, retrieval-augmented technology, or RAG, pairs an AI mannequin with a information base to offer the mannequin supplemental data earlier than it solutions, serving as a form of fact-checking mechanism.
Complete companies have been constructed on RAG, due to the sky-high demand for extra dependable AI. Voyage AI is one in all these. Based by Stanford professor Tengyu Ma in 2023, Voyage powers RAG techniques for firms together with Harvey, Vanta, Replit, and SK Telecom.
“Voyage is on a mission to enhance search and retrieval accuracy and efficiency in enterprise AI,” Ma advised TechCrunch in an interview. “Voyage solutions [are] tailored to specific domains, such as coding, finance, legal, and multilingual applications, and tailored to a company’s data.”
To spin up RAG techniques, Voyage trains AI fashions to transform textual content, paperwork, PDFs, and different types of information into numerical representations referred to as vector embeddings. Embeddings seize the that means and relationships between completely different information factors in a compact format, making them helpful for search-related purposes, like RAG.
Voyage makes use of a selected kind of embedding referred to as contextual embedding, which captures not solely the semantic that means of knowledge however the context wherein the info seems. For instance, given the phrase “bank” within the sentences “I sat on the bank of the river” and “I deposited money in the bank,” Voyage’s embedding fashions would generate completely different vectors for every occasion of “bank” — reflecting the completely different meanings implied by the context.
Voyage hosts and licenses its fashions for on-premises, personal cloud, or public cloud use, and fine-tunes its fashions for shoppers that decide to pay for this service. The corporate isn’t distinctive in that regard — OpenAI, too, has a tailorable embedding service — however Ma claims that Voyage’s fashions ship higher efficiency at decrease prices.
“In RAG, given a question or query, we first retrieve relevant info from an unstructured knowledge base — like a librarian searching books from a library,” he defined. “Conventional RAG methods often struggle with context loss during information encoding, leading to failures in retrieving relevant information. Voyage’s embedding models have best-in-class retrieval accuracy, which translates to the end-to-end response quality of RAG systems.”
Lending weight to these daring claims is an endorsement from OpenAI chief rival Anthropic; an Anthropic help doc describes Voyage’s fashions as “state of the art.”
“Voyage’s approach uses vector embeddings trained on the company’s data to provide context-aware retrievals,” Ma mentioned, “which significantly improves retrieval accuracy.”
Ma says that Palo Alto-based Voyage has simply over 250 clients. He declined to reply questions on income.
In September, Voyage, which has round a dozen workers, closed a $20 million Collection A spherical led by CRV with participation from Wing VC, Conviction, Snowflake, and Databricks. Ma says that the money infusion, which brings Voyage’s whole raised to $28 million, will help the launch of recent embedding fashions and can let the corporate double its measurement.
