Leading developer of GenAI hardware and software

Summary

Centific assured strict adherence to safety and governance policies for a leading AI assistant product team’s foundation model through prompt authoring and multi-turn red teaming, unifying more than 12 domains in one customized interface.

Challenge

To achieve the organization’s goals, the leading AI assistant product team needed its foundational model to strictly adhere to safety and governance policies. 

Solution

Centific:

• Built a dedicated team of linguistic specialists with backgrounds in varying domains to author prompts with the intention of deviating the model to identify areas for improvement.
• Developed a customized task interface to track safety policy violations complete with dashboards that reported on regressions and improvements with each model iteration.
• Curated a set of taxonomies aligned with the client's safety and governance policies. 

Results

This project resulted in immeasurable improvements to the safety and reliability of the client’s AI assistant capabilities, as well as a greater understanding of the harm potential of various models and applications.