Today, Montana-based data-as-a-service and cloud storage provider Snowflake introduced Cortex, a fully managed service that integrates large language models (LLMs) into its data cloud.
Unveiled at the annual Snowday event, Cortex equips enterprises using the Snowflake data cloud with a set of AI tools, including open-source LLMs, designed to analyze data and develop applications for various business-specific use cases.
“With Snowflake Cortex, businesses can quickly leverage large language models, build custom LLM-powered applications in minutes, and maintain flexibility and control over their data—transforming how users harness generative AI for business growth,” said Sridhar Ramaswamy, SVP of AI at Snowflake.
Available today in private preview, Cortex includes a suite of task-specific models to enhance specific functions within the data cloud. Snowflake plans to utilize this service for three of its generative AI tools: Snowflake Copilot, Universal Search, and Document AI.
Building LLM Applications with Cortex
While enterprises are eager to adopt generative AI, many struggle with the complexities of technology implementation, including the need for specialized talent and managing intricate GPU infrastructures. Snowflake Cortex aims to simplify this process.
The service offers users a range of serverless AI functions, both specialized and general-purpose. These functions can be accessed using simple SQL or Python calls, enabling users to initiate functional AI applications on Cortex's cost-optimized infrastructure.
Snowflake Cortex Architecture
The specialized functions utilize language and machine learning models to facilitate specific analytical tasks with natural language inputs. For example, these models can extract answers, summarize data, and translate languages. They can also assist with forecasting based on provided data or detect anomalies.
On the other hand, general-purpose functions provide developers with a diverse range of models, from open-source LLMs like Llama 2 to Snowflake's proprietary models, including one that converts text inputs into SQL for querying data.
Crucially, these general-purpose functions include vector embedding and search capabilities, allowing users to contextualize responses based on their data and create customized applications for various use cases, facilitated through Streamlit in Snowflake.
“This is fantastic for our users because they don't need to provision any resources,” Ramaswamy stated. “We handle provisioning and deployment, much like an API, similar to offerings from OpenAI, but entirely within Snowflake. Our customers can rest assured that their data remains isolated and secure, free from cross-customer training risks.”
Ramaswamy highlighted that extensive programming skills are not required; users can operate within SQL to accomplish their tasks.
On the application side, users can easily develop conversational chatbots tailored to their unique business knowledge, such as a copilot designed for help content.
Native LLM Experiences Powered by Cortex
Though Cortex has just been announced for enterprise usage, Snowflake is already enhancing its platform with three Cortex-driven features in private preview: Snowflake Copilot, Universal Search, and Document AI.
Snowflake Copilot serves as a conversational assistant that allows users to pose questions about their data in plain language, construct SQL queries, refine those queries, and extract insights.
Universal Search incorporates LLM-powered search capabilities, helping users quickly identify valuable data and applications relevant to their needs.
Document AI assists in extracting information from unstructured documents stored in the Snowflake data cloud, such as invoice amounts or contractual terms.
Other players in the data industry, such as Databricks, have introduced similar capabilities, including the recent launch of LakehouseIQ, enhancing competition with Snowflake. Informatica and Dremio also offer LLM solutions, enabling enterprises to manage and query data using natural language.
Additional Announcements from Snowday 2023
Beyond Cortex, Snowflake announced advancements in Iceberg Tables, allowing users to eliminate data silos and unify their information within the data cloud. The company also unveiled new features for its Horizon governance solution, including data quality monitoring, enhanced data lineage visualization, improved classification, and a trust center for cross-cloud security and compliance.
Lastly, Snowflake introduced a funding initiative aimed at investing up to $100 million in early-stage startups developing Snowflake-native applications, backed by its venture capital arm and several firms, including Altimeter Capital, Amplify Partners, and Menlo Ventures.
Ranking
