History 30/130A: The French Revolution: ChatGPT

It's great that ChatGPT can provide you with information - it's almost like a search engine, but remember that its knowledge is not current and that it's generating new text, so responses won't always be the same. 

It's important for you to critically evaluate the responses that you get from ChatGPT. According to OpenAi, there are three limitations you should know about ChatGPT:

• ChatGPT may create "hallucinations" - incorrect or misleading information that may sound convincing. Sometimes, these are facts that it makes up, but they can also be false citations or quotes. It might also take sides on an argument when you don't want it to.
• It may not have the most up-to-date information since its training data ends at September 2021. It also was trained primarily on English-language sources. So it knows quite a bit, but it doesn't know everything. 
• Additionally, it doesn't have access to search online to confirm results, nor can it verify facts or calculations.

While this list is mostly for ChatGPT-3.5, you should evaluate responses you get from any generative AI tool (Bard, Bing, ChatGPT-4, etc.). One of the most important skills moving forward will be the ability to critically evaluate information, so make sure to brush up on the technique presented below. Feel free to share any other techniques you use to validate information!

Since you want to evaluate the responses we receive from ChatGPT, you can use any of the techniques you've learned in class to evaluate your sources. However, there is another technique that you might find useful that you might not have heard about. 

This technique is called Lateral Reading and is similar to what fact-checkers do. Your goal when doing lateral reading is to look for other examples of the fact/statement that you are concerned about. This could involve checking references and confirming the information across multiple sources. For example, you may end up on several news sites to confirm that the event you looked up actually happened. 

For a more in-depth tutorial on lateral reading, check out this video from Crash Course Navigating Digital Information.

What are some other techniques you could think of to evaluate sources? Share them with others on the padlet below.

Below are some tips to help you develop useful prompts when working with different generative AI tools.

• Create Effective Prompts
  • Provide important context or information in your original prompt - this context will help the tool figure out the best response to provide you.
  • Use progressively detailed prompts. You can start with a general prompt, then gradually add additional details in follow-up prompts to get the results you are looking for
  • Similar to the two above, provide step-by-step instructions with each prompt to guide the tool
• Use Personas
  • Giving ChatGPT a persona allows it to answer your questions better. When provided a persona, ChatGPT will know how to answer the question from that perspective.
    • A persona is a set of traits or characteristics that includes tone, knowledge, perspective, etc. 
  • Examples of personas include:
    • "You are a college student..."
    • "You are an expert in [subject]..."
    • "You are a journalist writing for..."
• Incorporate user-provided examples
  • Share examples of desired output with your prompt - this gives ChatGPT guardrails on how to best respond and match your expectations.