Snowflake infuses enterprise AI throughout with Cortex AI and Snowflake ML advances

Snowflake began life as a cloud-native data warehouse product, before adding more and more capabilities such as secure data sharing, that saw it adopt the tagline "the data cloud." Well, now it's the "AI data cloud", with the product receiving more and more AI throughout. And, not simply AI that you can apply to your data such as training LLMs, but AI embedded into the product itself.

Snowflake's AI smarts, Snowflake Cortex AI, has been enhanced making it simple to create custom chat experiences, fine-tune best-in-class models, and expedite no-code AI development through new Cortex Search and Cortex Analyst, with other AI enhancements including Snowflake Copilot and Snowflake AI & ML Studio. Snowflake ML has been augmented with MLOps helping teams to seamlessly discover, manage, and govern their features, models, and metadata across the entire ML lifecycle.

With each of these, Snowflake has been careful to ensure outputs are trustworthy and reliable, and importantly, operate under the same security and governance as the rest of Snowflake so that any information is as protected through AI tools as it is within the rest of the product.

{loadposition david08}

“Snowflake is at the epicentre of enterprise AI, putting easy, efficient, and trusted AI in the hands of every user so they can solve their most complex business challenges, without compromising on security or governance,” said Snowflake head of AI Baris Gultekin. “Our latest advancements to Snowflake Cortex AI remove the barriers to entry so all organizations can harness AI to build powerful AI applications at scale and unlock unique differentiation with their enterprise data in the AI Data Cloud.”

New chat experiences

While the promise of AI has been with us for decades, waiting for computing power to catch up with science fiction, there's no denying generative AI has absolutely thrust AI into the public conscience and is already dramatically transforming the way we work. Of course, at the same time, it's well known that ChatGPT hallucinates - making claims that, in one case, saw lawyers fined for submitting cases and quotes as part of their legal submission which never existed and were purely a creation of ChatGPT.

ChatGPT is fun for creative writing, for poetry, for idea generation, but if you want a tool that allows your business users to ask questions of data in natural language and get trustworthy, accurate, secure, non-data-breaching results, you need enterprise-grade LLMs and enterprise-grade tooling. And that sounds like hard work.

iTWire already spoke about Snowflake's own LLM, Arctic, and Snowflake has also announced two new chat capabilities, Snowflake Cortex Analyst and Snowflake Cortex Search, which will both be in public preview soon.

These tools will allow Snowflake users to deliver chatbots in literally seconds - in fact, during Snowflake Summit 2024, EVP of product Christian Klienerman selected a random audience member to work through the easy point-and-click wizard to demonstrate its simplicity. Well, there was some pre-preparation - creating the semantic model, for example, that defines what the data represents and how it interrelates - but for the most part, Snowflake has abstracted away complexity to give users a user interface that leads through the steps to make a chatbot.

The chatbot can work with structured and unstructured data and Snowflake assures no data leaks to the Internet because the model runs on the same Snowflake cloud infrastructure where data is stored. And, the data will respect all the rules you have put in place for your data, such as row-level security or data masking, to control what specific users are allowed to see. Further, the bot will provide referenceable sources for the output it gives and will advise if it does not know an answer.

Cortex Analyst, built with Meta’s Llama 3 and Mistral Large models, allows businesses to securely build applications on top of their analytical data in Snowflake. In addition, Cortex Search harnesses state-of-the-art retrieval and ranking technology from Neeva alongside Snowflake Arctic embed so users can build applications against documents and other text-based datasets through enterprise-grade hybrid search — a combination of both vector and text — as a service.

Snowflake acquired Neeva in May 2023, and not only did it integrate the Neeva tech into its own platform, Neeva CEO Sridhar Ramaswamy became the Snowflake CEO three months ago.

What's more, data security is key to building production-grade AI applications and chat experiences, and then scaling them across enterprises. As a result, Snowflake also announced Snowflake Cortex Guard, to be generally available soon, which leverages Meta’s Llama Guard. That product is an LLM-based input-output safeguard that filters and flags harmful content across organisational data and assets, such as violence, hate speech, self-harm, or criminal activities. With Cortex Guard, Snowflake is further unlocking trusted AI for enterprises, helping customers ensure that available models are safe and usable.

Snowflake strengthens AI experiences to accelerate productivity

Snowflake is also providing customers with pre-built AI-powered experiences, fueled by its world-class models. One of these is Document AI, generally available soon. This product allows users to easily extract structured data out of documents - literally, PDF documents such as contracts, or scans of paper-based forms with handwritten scrawl such as machine inspection checklists or more. It's a remarkable tool which genuinely, seriously, means your mass and mess of filing cabinets can become a rich source of data without anyone needing to re-key them. Document AI was first announced last year, and iTWire was impressed then and sees this as bringing incredible value to organisations around the world.

Document AI is based on Snowflake's multimodal LLM Snowflake Arctic-TILT, which outperforms GPT-4 and secured a top score in the DocVQA benchmark test — the standard for visual document question answering. Snowflake is also advancing its breakthrough text-to-SQL assistant, Snowflake Copilot - generally available soon - which combines the strengths of Mistral Large with Snowflake’s proprietary SQL generation model to accelerate productivity for every SQL user.

Snowflake co-founder and president of product Benoit Dageville emphasised that a key tenet of Snowflake is to make everything it does available all throughout the product, and Document AI is no exception. Users will be able to call Document AI via SQL queries, or as part of other programmatic interfaces, rapidly converting paper-based documents into structured information for action and insights.

Unlock no-code AI development with the new Snowflake AI & ML Studio

Snowflake Cortex AI provides customers with a robust set of state-of-the-art models from leading providers including Google, Meta, Mistral AI, and Reka, as well as, of course, Snowflake Arctic, to accelerate AI development. Snowflake is further democratising how any user can bring these powerful models to their enterprise data with the new Snowflake AI & ML Studio, currently in private preview. This is a no-code interactive interface for teams to get started with AI development and productise their AI applications faster. In addition, users can easily test and evaluate these models to find the best and most cost-effective fit for their specific use cases, ultimately accelerating the path to production while optimizing operating costs.

To help organizations further enhance LLM performance and deliver more personalized experiences, Snowflake is introducing Cortex fine-tuning, now in public preview, accessible through AI & ML Studio or a simple SQL function. This serverless customisation is available for a subset of Meta and Mistral AI models. These fine-tuned models can be easily used through a Cortex AI function, with access managed using Snowflake role-based access controls.

Streamline model and feature management with unified, governed MLOps through Snowflake ML

Once ML models and LLMs are developed, most organisations struggle with continuously operating them in production on evolving data sets. Snowflake ML brings MLOps capabilities to the AI Data Cloud, so teams can seamlessly discover, manage, and govern their features, models, and metadata across the entire ML lifecycle — from data pre-processing to model management. These centralised MLOps capabilities also integrate with the rest of Snowflake’s platform, including Snowflake Notebooks and Snowpark ML for a simple end-to-end experience.

Snowflake’s suite of MLOps capabilities includes the Snowflake Model Registry, now generally available, which allows users to govern the access and use of all types of AI models allowing them to deliver more personalised experiences and cost-saving automations with trust and efficiency. In addition, Snowflake is announcing the Snowflake Feature Store, now in public preview, an integrated solution for data scientists and ML engineers to create, store, manage, and serve consistent ML features for model training and inference, and ML Lineage, currently in private preview, so teams can trace the usage of features, datasets, and models across the end-to-end ML lifecycle.

Continued Innovation at Snowflake Summit 2024

Snowflake also announced new innovations to its single, unified platform that provide thousands of organizations with increased flexibility and interoperability across their data; new tools that accelerate how developers build in the AI Data Cloud; a new collaboration with NVIDIA that customers and partners can harness to build customized AI data applications in Snowflake; the Polaris Catalog, a vendor-neutral, fully open catalogue implementation for Apache Iceberg; and more during its annual Snowflake Summit 2024 this week in San Francisco.