Working with Generative AI

Understanding and utilizing Generative AI as a Personal Knowledge Management (PKM) tool can be a useful way to using applications like ChatGPT, Claude, Bard, and others. There are two fundamental keys to success:

  • asking good questions, and
  • engaging in lengthened exchanges with the AI.

These interactions are not only about getting answers; they're about generating ideas, creating content, and making informed decisions, which you'll further enrich with credible information from library and web sources later in the semester.

In this section, we'll delve into the science of Prompt Engineering. This is the act of crafting your questions and commands in a way that guides the AI to understand and respond with the most useful and relevant information. A well-constructed prompt can be the difference between getting a generic response and unlocking valuable insights.

Reading

Below I demonstrate a session with Google Bard where I construct a series of prompts for Bard to talk about prompt engineering. The goal of the session was to show how using generative AI effectively is iterative, which means that I apply answers from Bard to generate new responses.

This session is based on readings from several sources. Before reading through the session, please review the following guides on prompt engineering tactics and techniques:

The above link outlines the importance of the following in interacting with generative AI:

  1. Write Clear Instructions: The clearer your request, the better the AI can meet your needs.
  2. Provide Reference Text: Supplying context or examples can significantly steer the AI towards the kind of response you're looking for.
  3. Split Complex Tasks into Simpler Subtasks: Breaking down bigger challenges makes them more manageable for the AI.
  4. Give Generative AI Models Time to Think: Sometimes, complex queries require a moment for the AI to process and respond thoughtfully.
  5. Using External Tools: Learn how to integrate other tools and resources to augment the AI's capabilities.
  6. Testing Changes Systematically: Experiment and evaluate how different approaches impact the AI's responses.

The above site provides important information on prompting. The Examples of Prompts page covers the following topics on ways to use generative AI:

  • text summarization
  • information extraction
  • question answering
  • text classification
  • conversation
  • code generation
  • reasoning

In the Techniques section, the Prompt Engineering Guide describes multiple ways to engage with generative AI. The first few items listed are good entry level approaches to construct prompts. These include:

  • zero-shot prompting
  • few-shot prompting
  • chain-of-thought prompting

And more. Please read through these techniques.

Prompt Exchange with Google Bard

The following is a link to a session I conducted with Google Bard. In this session, I prompt Bard about prompt engineering, and then work with Bard to provide some examples that demonstrate few-shot prompting and chain-of-thought prompting. I then ask Bard to show I could have improved my prompts, and it offers some suggestions. See the link below to read through the session:

Google NotebookLM

One of the downsides right now with generative AI clients is that most applications do not integrate a work flow for using sessions with AI into creating outputs, like essays. This is what something like Google's new NotebookLM seeks to solve.

For example, in the prior session with Bard, I asked Bard if it would offer suggestions for improving my prompts in a use case involving using generative AI to develop a strategy for writing an essay on the ethics of generative AI. Bard offered the following advice:

  • add specificity within steps
  • provide examples and case studies
  • provide alternate perspectives
  • prompt for creative solutions

To apply these suggestions using NotebookLM, I can add my own documents and data. NotebookLM lets users create topical notebooks and then add data from their Google Drive accounts, PDFs, and from copied text. By doing this, I would be able to address Bard's suggestions to provide examples and case studies and to provide alternate perspectives on a topic as I collect information from the library or the web.

Conclusion

Generative AI is a fairly new technology that offers a lot of potential for personal knowledge management. In this lesson, we learned how to use generative AI effectively by engaging in lengthy sessions that involve creating well-constructed prompts. To construct good prompts, we employed tactics such as few-shot prompting and chain-of-thought prompting.

With new tools such as Google's NotebookLM, we learned how to incorporate our own data that we have collected to continue to work on our prompts and to generate new content, ideas, and perhaps even, knowledge.