Skip to Main Content

Getting Started with AI

Learn how to use generative AI effectively and safely.

Prompting AI

AI models generally have a prompt box where you type in your request. While you could give it a simple prompt like 'write me a story', it is unlikely the output will be what you want unless you don't care about what is generated. It is better to give specific details of your needs without going overboard.

What to include in a prompt:A staring robot with the caption: ChatGPTs face when I ask it to write an essay, make it sound like a first year English student and sound human

  • State the type of output you want
    • Present in the form of a short story
  • Use "do" and "don't" in instructions
    • Write a story about fruit bats, not vampire bats
  • Include examples
    • I want the story written in the style of this text I pasted here [Pasted context]
  • Mention tone and audience
    • Write the story in an informational tone for second graders

You don't have to include all these factors in your first prompt. You can build on previous prompts by correcting mistakes or giving feedback. If you are stuck you can ask the model to create a prompt for you by starting with a basic idea of what you want.

You can read more about AI prompts at Harvard University (https://huit.harvard.edu/news/ai-prompts).

You can view a list of finely tuned prompts at Hugging Face (https://huggingface.co/datasets/fka/awesome-chatgpt-prompts).

 

AI Hallucinations & Wrong Information

Since generative AI is just a fancy autocomplete feature, it is wrong a lot and needs to be fact-checked. Often citations or summaries are wrong, as AI has trouble with larger blocks of text and has no fact-checking features. This is called a hallucination.Screencap of Gemini telling a user to add glue to their pizza because cheese wasn't sticking.

As an example, one user on Twitter noticed that Google's Gemini was very wrong about keeping cheese stuck on pizza (as seen in the image on the right). It told him to add glue. He looked into it and found that the comment was actually from a Reddit post that was over a decade old. Gemini had fetched incorrect information a troll had left on Reddit, and used that to give its output.

Likewise, AI can and will spit out wrong information and sources. Do not trust generative AI to be a trustworthy source, but rather as a starting point in your information search.

What can AI do?

To be able to use AI efficiently, you need to know everything that it is capable of. Generative AI is capable of producing a wide range of outputs, from reports and videos to even interactive help.

It can create:

  • Reports
  • Videos
  • Emails
  • Excel sheets
  • Images
  • Songs
  • Code
  • Entire websites

It can also allow you to take pictures of your screen and environment and ask questions or be walked through a process, such as finding settings on a printer.

However, a lot of features are locked behind the paywalls of subscriptions.

Digital Divide

The digital divide is a term used to explain the divide between users who can pay for the higher tier models and those who cannot. Those who pay for access can have AI general entire completed projects at just the click of a button while those using the free versions will have to painstakingly craft documents step-by-step and will have to use multiple tools for the same result the paid user gets with a single click.

So while generative AI is amazing and can produce many things, only those who pay for it will be able to take full advantage of the technology which creates a two tiered opportunity.