💻 From Research to Production: AI Model Safety in Practice

We will talk about how to safely deploy AI models in production. There are a number of risks that originate from deploying various closed source or open source models that can cause businesses to lose customers or affect the brand name negatively. For example: Models may produce unsafe outputs, they may recommend competitor brands, they may talk openly about scandals related to the company that is deploying the model etc. The talk will provide a framework on which risks to consider, how to build an evaluation framework to measure these risks and alignment strategies to mitigate these risks before deployment.