[San Francisco (USA) = Reporter Kim Uyong] On the second day of the ‘Snowflake Summit 2024’ held in San Francisco, USA on the 4th, ‘Snowflake Cortex AI’ announced new features.
Snowflake Cortex AI is a tool that easily and efficiently generates AI-based applications, allowing companies to create chatbots within minutes and obtain answers utilizing their own data in chat environments through the latest updates.
Snowflake’s new no-code interactive interface and access to large language models (LLM) enable non-experts to easily utilize AI tailored to specific industry use cases, promoting ‘data democratization’. With Snowflake ML, a unified environment for machine learning is provided, enabling rapid development of routes to operate models. Snowflake’s generative AI and ML integrated platform generates more value from data across businesses, offering reliable AI at scale with complete security, governance, and control.
The strength of LLM-based chatbots is that non-experts can utilize internal company data through natural language queries. Companies can leverage this to quickly gain insights necessary to make important decisions.
Snowflake introduced two new chat features, ‘Snowflake Cortex Analyst (public preview scheduled)’ and ‘Snowflake Cortex Search (public preview scheduled)’. Users can develop chatbots based on their structured and unstructured data without complex procedures in just a few minutes.
With Cortex Analyst built on Meta’s LAMA3 and Mistral Large models, companies can safely and easily create generative AI-based applications. Cortex Search leverages advanced search and ranking technology acquired from Neeva in May 2023, combined with Snowflake Arctic Embed, to provide enterprise-level hybrid search combining vector and text. This enables building applications for documents and other text-based datasets. Even non-experts can quickly set up AI-based chat services through simple steps.
Snowflake introduces Cortex Guard utilizing Meta’s LAMA Guard, an LLM-based input/output protection feature that filters harmful content such as violence, hate, self-harm, or criminal activities across enterprise data and assets. Through Cortex Guard, Snowflake enhances AI safety for enterprises.
Snowflake provides not only customized chatbot development but also pre-built AI-based experiences. Users can easily extract content such as invoice amounts or contract terms using DocumentAI with Snowflake’s multimodal LLM ‘Snowflake Arctic TILT’. This model outperforms GPT-4 and achieved the highest score in the DocVQA benchmark test for visual document question-answering standards.
Companies like Northern Trust are reducing operating costs and increasing efficiency by processing documents in bulk using DocumentAI. Snowflake will also release Snowflake Copilot, a text-to-SQL assistant that combines Mistral Large’s strengths with Snowflake’s proprietary SQL generation model to enhance the productivity of all SQL users.
Snowflake Cortex AI accelerates AI development by providing customers with state-of-the-art models from Google, Meta, Mistral AI, Leka, and other reputable companies in addition to Snowflake’s top-tier open-source LLM, Snowflake Arctic. Snowflake democratizes data utilization through the new Snowflake AI & ML Studio (private preview), a no-code interactive interface supporting enterprises in applying their own data easily, allowing users to select the most suitable model through simple testing and evaluation processes, ultimately optimizing operating costs and shortening the period to production.
Snowflake releases Cortex Fine-Tuning (public preview) to enhance LLM performance and provide a more personalized experience, offering a serverless customization feature provided as part of Meta and Mistral AI models. Refined models can be easily accessed through Cortex AI features, managed with Snowflake role-based access control.
Most companies struggle to continuously operate ML and LLM models in changing datasets after development. Snowflake ML provides ML Ops capabilities in AI data clouds, enabling organizations to seamlessly discover, manage, and control features, models, and metadata across the entire ML lifecycle from data preprocessing to model management. These centralized ML Ops capabilities integrate with Snowflake platforms including Snowflake notebooks and Snowpark ML to provide an easy end-to-end experience.
Snowflake’s ML Ops capabilities include the Snowflake Model Registry which controls access and usage of all types of AI models, ensuring reliability, efficiency, personalized experiences, and cost-saving automation. Snowflake announces the Snowflake Feature Store (public preview) and ML Lineage (private preview), integrated solutions enabling data scientists and ML engineers to create, store, manage, and serve consistent ML features, datasets, and model usage traceability.
Chris Kleyneman, Snowflake’s EVP, said, “Snowflake’s AI services empower everyone in the enterprise to access data and provide huge opportunities.” He added, “Cortex AI is triggering multiple innovations by making AI development workflows, which consist of multiple complex steps, easy to build.” He emphasized, “With the power of data and AI, we can do more,” promising to help organizations lead the enterprise AI era, find more value and insights and continue to innovate constantly.