Generative AI (gen AI) can be transformative for businesses. Enterprises can utilize gen AI to extract more value from their data and build conversational interfaces for customer and employee applications. Adopting AI solutions can enhance overall efficiency. Gen AI enables employees to perform enterprise-grade tasks with greater speed and accuracy, driving productivity and engagement. Additionally, upon implementing robust data security controls and meeting regulatory requirements, businesses can confidently integrate AI while meeting compliance standards. Addressing a lack of in-house AI expertise and simplifying AI processes can make adoption easier. By taking advantage of these opportunities, enterprises can use gen AI to achieve greater success. 

That’s where Snowflake comes in. Snowflake Cortex AI is a fully managed service designed to unlock the potential of the technology for everyone within an organization, regardless of their technical expertise. It provides access to industry-leading large language models (LLMs), enabling users to easily build and deploy AI-powered applications. By using Cortex, enterprises can bring AI directly to the governed data to quickly extend access and governance policies to the models.  

We recently announced updates to Cortex AI, including the general availability of LLM functions that provide serverless access to foundation and embed models. 

To further enable enterprises to make full use of gen AI in a way that is efficient, easy and trusted, we are introducing a new wave of components in Cortex AI, including: 

• Cortex Analyst (public preview soon): Enable business users to interact with structured data using natural language, allowing them to find answers faster, self-serve insights and save valuable time. 
• Cortex Search (public preview soon): Quickly and securely find information by asking questions within a given set of enterprise documents using the state-of-the-art Arctic embed model. 
• Cortex Fine-Tuning (public preview): Customize LLMs securely and effortlessly to increase model accuracy and performance for use-case specific tasks using serverless fine-tuning. Fine-tuned models can be managed using Snowflake Model Registry.
• Snowflake AI & ML Studio for LLMs (private preview): Enable users of all technical levels to utilize AI with no-code development. For ML model development, the Studio is in public preview.
• Snowflake Cortex Guard (generally available soon): Use an LLM-based safeguard to filter and flag harmful content, ensuring model safety and usability.

Other enhancements include adding REST APIs to programmatically integrate Cortex AI functionality in any application of your choice and the support for AI21 labs models (coming soon).

Below are the details for each of these updates.

Cortex Analyst, built with Meta’s Llama 3 and Mistral Large models, lets you get the insights you need from your structured data by simply asking questions in natural language. Cortex Analyst enables app developers to build applications for business users on top of analytical data stored in Snowflake. Developers can build conversational data applications powered by AI that break down data silos and give everyone in an organization access to quick answers and insights from its data.

Cortex Analyst is designed to help any user find answers inside Snowflake. It helps your teams work quicker and more efficiently by empowering self-service data insights and decision-making. To deliver highly precise results that business teams can trust, Cortex Analyst uses semantic models tailored to your business to understand the specific terms and data structure used within the context of an organization — resulting in more accurate and relevant results. 

Snowflake Cortex Analyst is designed to let you ask questions from any application that you’re interacting with to get the answers you need. It serves another end of the business, as compared to the Snowflake Copilot assistant, which helps SQL developers accelerate development from inside the Snowflake UI by turning text into SQL. 

Cortex Search is a fully managed search solution that offers a rich set of capabilities to index and query unstructured data and documents. Cortex Search manages the end-to-end workflow for data ingestion, embedding, retrieval, reranking and generation. The embedding process is completely automated for easy configuration. Our state-of-the-art hybrid search enables better results.

Cortex Search simplifies the implementation and integration of search in your applications. Specifically, it offers:

1. Integrated document preprocessing: To streamline the process of extracting text from files (pdf-to-text) and splitting text into chunks before indexing, Snowflake Cortex will be adding two managed functions in private preview soon. With such preprocessing, developers won't have to implement or deploy custom functions required to have text ready before embedding. 

2. Fully managed vector embedding and management: With Cortex Search, there is no separate API or service needed to generate vector embeddings. This removes the overhead required to turn text into vector format. Using Snowflake data processing infrastructure, the service is kept up-to-date with the latest information by automating continuous refreshes as new documents are generated.

3. State-of-the-art search and ranking: To provide more accurate results, Cortex Search uses state-of-the-art retrieval and ranking techniques. Using a combination of semantic and keyword search, built on the cutting-edge technology from Neeva and the Arctic embed models built by the Snowflake AI Research team, Cortex Search is able to provide users with high-quality results without operational complexity. Behind the scenes, Snowflake Cortex conducts searches in parallel, merging and ranking the results to provide accurate outputs grounded on the relevant document information. 

4. Security and governance: Because Cortex Search is integrated into Snowflake’s platform, consistent governance policies can apply to both the raw documents and vector embeddings. This minimizes data risk and reduces time spent maintaining separate data security frameworks. 

With fully managed vectors and retrieval, Cortex Search helps organizations streamline the Retrieval Augmented Generation (RAG) architectures used to bring private, up-to-date information to LLMs for more accurate results. In this model, Cortex Search first fetches relevant documents or passages from a knowledge source, based on the input question. Then, the relevant result set can be passed to an LLM in Snowflake Cortex to generate accurate and contextually relevant responses.

In addition to simplifying the development of RAG-based chatbots, Cortex Search can increase the productivity of teams that regularly look for information spread across large document corpuses. Specifically, you can perform: 

1. Needle-in-a-haystack lookups: Find the answer you need hidden in large amounts of documents.

2. Multidoc synthesis and reasoning: Source answers to questions that are strung across several documents. For example, you could ask, “How did earnings of my company grow between the years 2022 and 2023?”

3. Tabular data, figure synthesis: Get answers from non-free-text structures in your data — such as data specifically formatted and often found in common data storage systems, like databases and spreadsheets.

With Cortex Fine-Tuning, you can fine-tune industry-leading models from Meta and Mistral AI. Simply put, fine-tuning an AI model is when you make small adjustments to the model to improve its performance on a specific task or data set. This helps developers improve accuracy for specific tasks, such as summarization, or generate better outputs for the task they are working on. With Cortex Fine-Tuning, you can fine-tune a relatively smaller base model to achieve the same accuracy as that of a larger model. By doing so for oft-repeated, specialized tasks, you can achieve lower inference latency and lower costs.

With Cortex Fine-Tuning, you can fine-tune by calling an API or SQL function, all without the hassle of managing any infrastructure. Just select the base model and provide your training data set, and Cortex creates the fine-tuned model for you. Once the fine-tuned model is available, you can integrate it into your application. The fine-tuned models are secure because you can manage access to the models using Snowflake role-based access control (RBAC). 

Once the fine-tune job is complete, the model can be accessed via Cortex COMPLETE, just like any other LLM.

Snowflake is making AI accessible to users of all skill levels. With Studio, developers can easily try out LLMs, fine-tune models with their data and interact with documents, all through the console. Studio also enables expedited delivery of applications by improving developer productivity.

From inside the Snowflake AI Studio, users will be able to:

• Compare completions across multiple LLMs (private preview): Using the chat playground, business users can easily test and evaluate LLM models to find the best and most cost-effective fit for their specific needs. This helps users improve productivity by quickly identifying the right settings and the right LLM to use for their application. 
• Fine-tune LLMs with UI (private preview): Users will be able to fine-tune their models with an easy-to-use UI. Users can develop custom models either through Studio’s no-code click-through experience or, for those who prefer code, via the designated fine-tuning function.

Additionally, users can use a prebuilt chat interface to ask questions against existing sets of indexed documents built with Cortex Search.

Snowflake is launching Cortex Guard, a new feature that gives enterprises the ability to filter and flag harmful content to ensure model safety and usability using Meta's Llama Guard. We built Cortex Guard because customers told us that AI safety is critical for deploying AI applications into production. For example, enterprises want to protect their user-facing applications from potentially unsafe AI responses that impact their services and ultimately their business. The new Cortex Guard feature is purpose-built for these use cases. Cortex Guard works by evaluating the responses of a Language Model before that output is returned to the application. With a single click, customers can now enable the new Cortex Guard parameter available within the COMPLETE function.

With REST APIs (coming soon), you can programmatically access LLMs on Cortex from your applications. You can configure a prompt and run inference using one of the LLMs. Alternatively, you can select from the prebuilt functions to summarize text, identify sentiment or translate input.

Gen AI allows enterprises to make fuller use of their data and drive overall efficiency at a faster pace. Snowflake Cortex AI makes it easy for enterprises to leverage gen AI capabilities by chatting with their data and so much more. 

We’ve added the tools and infrastructure you need to make it easy for all your employees to build and deploy AI-powered applications in a secure environment. They can now drive better-informed decisions, optimize operations and deliver personalized experiences to your customers and stakeholders by building these AI-powered apps with Snowflake Cortex AI. We’ve not only made it easy but also efficient and secure for enterprises like yours to unlock the full potential of gen AI.

To help you get started with Snowflake Cortex AI, try one of these quickstarts:

• Batch text processing:
  • Summarize text in JSON format using Snowflake Arctic
  • Analyze customer reviews analytics using Snowflake Cortex AI
• RAG-based chatbots:
  • Build maintenance logs app using Snowflake Cortex AI
  • Create document-search assistant using Streamlit and Snowflake Cortex AI

To accelerate development for business and industry-specific use cases, Snowflake customers can access Cortex AI through our trusted partners. Click here to learn more about their solutions on the Snowflake Cortex webpage.