Generative AI & Trust Principles

Einstein Generative AI & Trust

• Salesforce prioritizes data security and accuracy in its Einstein generative AI, ensuring user data remains private.
• Agreements with LLM providers (like OpenAI) prevent private data from being retained by the providers.

Five Principles for Trusted Generative AI

• Accuracy: Model responses backed by explanations and sources; human review recommended.
• Safety: Detecting and mitigating bias, toxicity, and harmful content using industry-best techniques.
• Transparency: Models respect data provenance and use user data.
• Empowerment: AI augments user capabilities, promoting efficiency and purpose.
• Sustainability: Focus on appropriate model size and reduced carbon footprint.

Reviewing Generative AI Responses

• User responsibility for LLM-generated content shared with customers; content aligns with company values and brand.
• Key aspects for review: accuracy and safety.
• Accuracy: Checking factual accuracy; verifying sources to avoid "hallucinations" (inaccurate responses).
• Bias and Toxicity: Reviewing responses for potential bias against marginalized groups or harmful language.
• Options include editing responses or generating new ones if needed.

Explore the essential principles that govern trusted generative AI systems, including accuracy, safety, transparency, empowerment, and sustainability. This quiz will challenge your understanding of Salesforce's approach to data security and the responsible use of AI technology. Test your knowledge on how organizations can ensure trust in AI interactions.