Implementing responsible AI safety guardrails on a chatbot using the Google Gemini API.
This takes an existing chatbot running on Google Cloud with VertexAI. Replace with your own project info to run it your self.
Then, we use diagnostics to measure the baseline safety in several scenarios.
Next, we tune the chatbot to respond appropriately when it is asked an in-appropriate question.
Finally, we test it with a prompt to make sure it is performing as expected.