Generative AI

What is generative AI? 

Generative AI refers to artificial intelligence that generates new, unique content, including but not limited to text, images, videos, code, music and speech. ChatGPT, Google Gemini, and Microsoft Copilot are a few examples of generative AI that you may already be familiar with.

In contrast, another common form of AI you might encounter is called discriminative AI. Discriminative AI can classify or recognize patterns in existing data but does not create new data. Some forms of discriminative AI that you may have encountered before include e-mail spam filters, facial recognition, and systems that make personalized recommendations for content or products you might enjoy (for example, on social media or video streaming services). 

What can generative AI do?

A lot! Generative AI can create new text, images, video, and audio. It can help you generate ideas, edit your writing, answer your questions (but not necessarily correctly!), translate text from one language to another, and rephrase information to make it easier to understand. There are many helpful uses for generative AI, but there are also many tasks that AI is not well-suited for. 

Can I trust the answers I get from AI?

No, don't take AI responses at face value. AI responses should always be fact-checked and critically evaluated. Responses may include "hallucinations" (inaccurate or misleading information fabricated by AI), outdated information, or content that is biased, offensive, or harmful. 

Although generative AI does often give correct answers, it doesn't actually comprehend information and can't recognize misinformation or disinformation, so it sometimes generates correct-sounding responses that are actually inaccurate or biased. Large language models operate on probability, much like the predictive texting on a smartphone, and they are not capable of critical thinking. Some AI tools are connected to web search and can summarize search results, but they can't tell a credible source from an unreliable one, and they can't access information sources kept behind paywalls. Other tools are not web search enabled, so they can't access the most up to date data or show you where their information came from.

AI technology is developing quickly and developers are working to reduce bias in AI, but, at least for now, AI tools can't guarantee accurate or unbiased results. 

What are prompts and how can I design an effective prompt?

To use generative AI effectively, you'll need to become skilled at crafting efficient prompts.

Prompts are the input users enter into a generative AI tool to guide its output. Prompts can take the form of questions, commands, statements, or a combination of all three. Prompt engineering is the process of designing and optimizing prompts to successfully guide the AI tool to produce the desired outputs. Check out the video below for a quick intro to prompting AI.

Here are some additional tips for creating strong prompts:

• Be specific. Using clear, precise language with generative AI will get you better results. You can use a casual or formal tone, but be direct and avoid ambiguous language. Feel free to use subject-specific terminology. 
• Define roles. Treat the interaction like role-play. Assign the AI tool a role and identify the role you'll be playing in the conversation. For example: 
  • "Act as a college English professor. I am a student trying to improve the organization and clarity of my writing."  
  • "Act as an expert in [subject]. I am a college student who is new to this subject, and I want to better understand..."
• Set a goal. Be specific about what you would like the AI model to generate. For example:
  • "Recommend ways that I can strengthen the following argumentative essay. Include suggestions for organizing the ideas logically and persuasively, strengthening my argument, and making the wording clear and easy to understand. Also identify any errors in grammar, spelling, or punctuation."
  • "Make a list of 15 keywords I could use to find books and articles about..."
  • "Provide and explain 3 examples of..."
• Provide context and examples. Extra context can help guide the AI model toward your desired outcome and clarify any words with multiple meanings (for example, whether you're using "bank" to mean a financial institution or the land along the edge of a river). Providing details, descriptions, and examples of what you're looking for will strengthen the response you receive and help tailor the output to your preferences. For example, if you're asking AI to help you come up with research topic ideas, you might want to include the subject of the class and some concepts from the course that you're particularly interested in or share specific instructions from the assignment prompt like themes or page limits. 
• Follow up with clarifying questions or further instructions. It may be helpful to follow up on the AI tool's initial response with any questions you have about its answer. You can also tell the AI model what you do and don't like about its suggestions and ask it to try again. For example:
  • "How could the ideas you've generated for new museum displays be made more interactive?"
  • "Try again but this time focus on...and leave out..."
  • "I like number 2 and number 5. Give me more ideas like those."
• Try multiple iterations and compare responses. Try asking the AI tool the same question again and look for differences in the response. Phrase the question differently based on what you did or didn't like about its previous responses to get an answer better tailored to your preferences. You can also try approaching the topic from a different perspective. (For example, changing "Why is..." to "Why isn't...") This can highlight inconsistencies and help you detect bias in the answers. Over time, comparing answers to slightly different versions of the same questions will help you identify strategies for engineering more effective prompts.