STD_1000_GenAI.pdf

STATE OF CALIFORNIA

GENAI DISCLOSURE & FACTSHEET

STD 1000 (NEW 01/2024)

DEPARTMENT OF GENERAL SERVICES

PROCUREMENT DIVISION

Generative Artificial Intelligence (GenAI) Disclosure & Factsheet Bidder/Offer Information

Solicitation Number Bidder ID/Vendor ID (optional)

Business Name Business Telephone Number

Business Address City State Zip Code

GenAI Disclosure & F actsheet Will you be using or offering GenAI technology, model, or service (collectively, “system”)? Yes No (If No, skip to Signature section of this form.)

If yes, provide details regarding the GenAI system”). See GenAI Disclosure & Factsheet Definitions at the end of this form for more information.

Failure to disclose GenAI to the State and submit the detailed description may result in disqualification and may void any resulting contract.

1. GenAI Model Name, Version (including number of parameters)

2. Model Owner

3. Overview

4. Purpose

5. Intended Domain

6. Model Training Data

7. Model Information

STD 1000 (NEW 01/2024)

DEPARTMENT OF GENERAL SERVICES

PROCUREMENT DIVISION

8. Input and Outputs

9. Performance Metrics

10. Optimal Conditions

11. Poor Conditions

12. Bias

13. Test Data

Explain below how you are ensuring the GenAI system is not adversely affecting “decisions that materially impact access to, or approval for, housing or accommodations, education, employment, credit, health care, and criminal justice.” (AB 302, Department of Technology: High-Risk automated decision systems: inventory).

Signature

By signing this document, I certify that I have identified and disclosed, if any, all GenAI components in the proposed solution or service.

Signature Date

STD 1000 (NEW 01/2024)

DEPARTMENT OF GENERAL SERVICES

PROCUREMENT DIVISION

GenAI Disclosure & Factsheet Definitions Please use the following definitions to complete the GenAI Disclosure and Factsheet:

1. Model Name, Version & Number of Parameters:

• Definition: The unique identifier or name assigned to the specific GenAI model or service.

• Purpose: Allows users to refer to and distinguish between different GenAI models.

2. Model Owner

• Definition: The name of the organization or entity responsible for creating or deploying the GenAI model or service.

• Importance: Helps identify the source and accountability for the GenAI system.

3. Overview:

• Definition: A concise summary of the GenAI model’s purpose, functionality, and key characteristics.

• Role: Provides a high-level understanding for users and stakeholders.

4. Purpose:

• Definition: The intended use or goal of the GenAI model (e.g., image recognition, natural language processing, text summarization).

• Significance: Helps users assess whether the GenAI model aligns with their needs.

5. Intended Domain:

• Definition: The context, subject matter or domain for which the GenAI model is designed to operate effectively.

• Importance: Helps users determine if the GenAI model is suitable for their specific use case.

6. Training Data:

• Definition: Information used to train the GenAI model (e.g., labeled images, text corpora).

• Role: Influences the GenAI model’s behavior and performance.

7. Model Information:

• Definition: Details about the architecture, parameters, and configuration of the GenAI model.

• Relevance: Provides insights into how the GenAI model functions.

8. Inputs and Outputs:

• Definition:

o Inputs: The data or features provided to the model for prediction (e.g., images, text).

o Outputs: The GenAI model’s predictions or results (e.g., class labels, probabilities).

• Understanding: Crucial for integrating the GenAI model into applications.

9. Performance Metrics:

• Definition: Quantitative measures (e.g., accuracy, F1-score) used to evaluate the GenAI model’s performance.

• Assessment: Determines how well the GenAI model meets its intended purpose.

• Continuous Monitoring Plan: Establishes a plan for continuous monitoring and evaluation of the GenAI model’s performance.

STD 1000 (NEW 01/2024)

DEPARTMENT OF GENERAL SERVICES

PROCUREMENT DIVISION

10. Optimal Conditions:

• Definition: The ideal environment or context for the GenAI model to perform optimally.

• Contextual Guidance: Helps users achieve the best results.

11. Poor Conditions:

• Definition: Scenarios or conditions where the GenAI model’s performance may degrade.

• Risk Awareness: Alerts users to potential limitations.

12. Bias:

• Definition: Any systematic error or unfairness in the GenAI model’s predictions due to biased training data or design.

• Mitigation: Addressing bias is crucial for ethical and unbiased GenAI.

13. Test Data:

• Definition: Independent data used to evaluate the GenAI model’s performance after training.

• Validation: Ensures the GenAI model generalizes well to unseen examples.