anyscale.com/gen-ai-terms

Generative AI Supplement

Generative AI Supplement

This Generative AI Supplement (“GenAI Supplement”) is a supplement to the Anyscale Terms of Service ("TOS") and governs your use of Anyscale-hosted Generative AI Models (defined below) provided via the Anyscale Platform, including via the public Anyscale Endpoints Service (collectively with Generative AI Models, the “AI Services”). Terms used in this GenAI Supplement but not defined herein shall have the meaning provided to them in the TOS. The AI Services shall be deemed “Platform Services” under the TOS, and the Agreement, as defined in the TOS will be deemed to include this GenAI Supplement with respect to the AI Services.

Large language models (also known as “LLMs”) are a subset of machine learning models designed to take input and generate output (“Generative AI Models”). They are an experimental and rapidly evolving technology. If you use LLMs or other Generative AI Models provided by Anyscale, you acknowledge that they have many limitations, including:

• Their randomness – Generative AI Models are inherently random in the way they create outputs, and providing the same input twice likely won’t return the same output;

• Their unreliability – Generative AI Models are, by nature, designed to generate output (e.g., text) that appears to have been generated by a human - they are not designed to provide factually correct answers. As a consequence, the results you receive may sometimes generate information that may sound authoritative, but is factually incorrect;

• They may create identical or near-identical output for multiple users; and

• They are not designed to provide tailored expertise – e.g., they are not designed to provide legal, financial, or medical information or advice.

• Description of Services. The AI Services allow users to input queries ("Customer Inputs") via API and UI to selected third-party licensed machine learning models for processing ("Chosen Models"), receive responses to the Customer Inputs from each of the Chosen Models ("Model Outputs"), and generate metrics (e.g., latency, cost, tokens) for each of the Chosen Models (“Performance Metrics”). Users may also be provided a method to indicate which of the Model Outputs is superior ("Rankings") and may be permitted to store Customer Inputs, Model Outputs, and Rankings. The Customer Inputs, Model Outputs, Performance Metrics and Rankings are collectively referred to as “AI Data”; Customer Input and Model Outputs shall be considered “Customer Data” under the TOS.

• Third Party Products. Any third party software, services, or other products you use in connection with the AI Services, including without limitation the Chosen Models (“Third Party Products”), are subject to their own terms and conditions and Anyscale is not responsible for Third Party Products. You expressly acknowledge and agree that your use of any Model Outputs may be subject to terms imposed by the providers of the Chosen Models. Anyscale makes no representations or warranties regarding, any Third Party Products, and your access to and use of such Third Party Products are at your own risk, which risk you expressly assume.

• Deployment of Generative AI Models in Dedicated Compute. You acknowledge that you have a choice to use the Anyscale Platform to deploy Generative AI Models using dedicated (i.e., non-shared) compute resources (deployed either into your Customer Data Plane or the Anyscale Data Plane) rather than the Anyscale-hosted public Anyscale Endpoints service. If you choose to use the Anyscale-hosted public Anyscale Endpoints service, including within the Anyscale Platform, and notwithstanding your choice of Cloud Service Provider for your dedicated compute resources, Anyscale may select one or more of the Cloud Service Providers listed here at our sole discretion and without regard to the nature or sensitivity of your Customer Data. Notwithstanding anything to the contrary in the TOS or the DPA (including without limitation Section 3.6(b)), you expressly acknowledge and agree that: