Class 5
beyond chatbots
1. Introduction
How can we tailor our chatbots and how can we use different modalities to achieve our goals? In this video, we will look at examples from healthcare and education to understand when and how to use tailoring to make our AIs more effective.
2. System Prompts
Let's explore different approaches to tailoring: system prompts, retrieval augmented generation (RAG) and fine-tuning. In this video, we focus on system prompts as a way to allow users to give chatbots a context that will be used for every interaction. We test this strategy with an example using OpenAI's Custom GPTs.
3. RAG and fine-tuning
In RAG, we provide our chatbot with additional sources of information that it can use to come up with an answer to our queries. In fine-tuning, we use comparison data to help our chatbot understand what we are looking for. In this video, we use OpenAI's Custom GPTs to test the RAG approach.
📝 Your turn! Create a RAG-based AI assistant for the HKS Admissions Office, as instructed in the video. You can use this background information. Note: you will need to purchase a ChatGPT Plus account for this activity, and you can do so here.
4. Custom GPTs
How did your bot do? What tasks is the bot particularly good or bad at? In this video, we brainstorm on some of the lessons learned.
5. Modalities beyond Chatbots
In this video, we focus on two main modalities: AIs integrated in user interfaces and AIs using formats other than text. We'll look at a demo of Microsoft Copilot's ability to create entire slide decks, and we'll test a data analysis scenario using ChatGPT.
6. Vision and Key Takeaways
Have you ever found yourself staring at your fridge, trying to come up with a good recipe for your meal? We certainly have, so we decided to ask ChatGPT to help. In this video, we use this example to explore AI's vision capabilities.