Snowflake AI and ML

Snowflake offers two broad categories of powerful, intelligent features based on Artificial Intelligence (AI) and Machine Learning (ML). These features can help you do more with your data in less time than ever before.

  • Snowflake Cortex is a suite of AI features that use large language models (LLMs) to understand unstructured data, answer freeform questions, and provide intelligent assistance. This suite of Snowflake AI Features comprises:

  • Snowflake ML provides functionality for you to build your own models.

    • ML Functions simplify the process of creating and using traditional machine learning models to detect patterns in your structured data. These powerful out-of-the-box analysis tools help time-strapped analysts, data engineers, and data scientists understand, predict, and classify data, without any programming.

    • For data scientists and developers, Snowflake ML lets you develop and operationalize custom models to solve your unique data challenges, while keeping your data inside Snowflake. Snowflake ML incorporates model development classes based on popular ML frameworks, along with ML Ops capabilities such as a feature store, a model registry, framework connectors, and immutable data snapshots.

Use of Snowflake AI Features

Snowflake AI Features and their underlying models are designed with the following principles in mind:

  • Full security. Except as you elect, all AI models run inside of Snowflake’s security and governance perimeter. Your data is not available to other customers or model developers.

  • Data privacy. Snowflake never uses your Customer Data to train models made available to our customer base.

  • Control. You have control over your team’s use of Snowflake AI Features through familiar role-based access control.

AI/ML model update process

Snowflake is continually working to improve the quality of its offerings, including the models powering the Snowflake AI Features. This section describes how updates to those models fit into Snowflake’s Behavior Change process.

Behavior change process for models

At Snowflake, feature updates are announced and deployed in the following 3 types of releases:

Model updates follow a similar pattern of announcements. For model updates, the following would constitute a behavior change:

  • Required syntax changes (e.g. specifying a new model or model version in the function parameter).

  • Required prompts or input updates to get similar results.

  • Significant changes in structure of the model output.

  • Deprecation of a model.

Bundled behavior changes would include most anticipated behavior changes, including:

  • Model deprecation in the ordinary course, such as planned deprecation by the model provider or Snowflake (including those on which fine-tuning is permitted).

  • Model updates, e.g. new versions or new models, that may result in changes to syntax, prompts, or output structure.

Unbundled behavior changes would typically be reserved for the following:

  • Model deprecation for emergency reasons, e.g. concerns about the quality of a model or its outputs.

Lastly, What’s new denotes general improvements that would likely not constitute a behavior change and therefore would be automatically included. This would typically be the following:

  • Model updates or new versions (whether provided by a third party or Snowflake) that improve results but have no anticipated material effect on how you interact with the model.

The following table shows some examples of model updates and how they would be announced:

Type of update

Unbundled behavior change

Bundled behavior change

What’s new

A new version of the Jamba model is released but has no anticipated material effect on how you interact with the model.

A new Llama model is made available through Snowflake.

One of the Mistral models is deprecated.

An update to the TRANSLATE model results in a change in the output structure.

A model is deprecated due to safety concerns regarding the model output.

Language: English