ALC Workshop - Building an AI Red Teaming tool

Just as with cyber security, we need to proactively test AI applications to ensure they are secure prior to launching them into production. We also need to make sure the Large Language Models we use do not generate biased, toxic, harmful, or hallucinatory responses and do not expose sensitive data. 

AI Red Teaming is a form of vulnerability assessment and penetration test which we can employ for proactive testing. In this workshop, we will design and build an AI red teaming platform using open source components, use AI models to help generate a target-focused test dataset, and use our tool to test some current huggingface models. We'll use the Google Colab environment and build the tool as a python notebook during the workshop, and we'll conclude by learning how to download it and run it as a local webapp on own laptop.