It is ironic that much of the early conversation about ChatGPT had to do with the demise of writing skills when, in fact, the very success of generative AI searches is determined by the quality of precisely-written prompts.

Let’s begin with the term “generative AI.” McKinsey Explainers concisely describes it this way: “Generative artificial intelligence (AI) describes algorithms (such as ChatGPT) that can be used to create new content, including audio, code, images, text, simulations, and videos. Recent breakthroughs in the field have the potential to drastically change the way we approach content creation.” In common practice, many simply generalize all such apps under the single name of ChatGPT, the application developed by the non-profit, although heavily funded by Microsoft, OpenAI. Yet, they include Bard, Perplexity.ai, You.com, Bing Chat and a host of other apps. In all cases, however, they all require an input in the form of a prompt that describes what task the application is to perform. It is the quality and detail of the prompt that determines the quality and relevance of the output.

The writing of prompts is both an art and a science. It requires you to write as if you were writing, or speaking, to an intelligent colleague, while taking little shared background for granted. David Gewirtz of ZDNet writes “When talking to a person, it would be natural to expect someone to miss your point initially and require clarification. It would be natural for someone you’re talking to veer away from the topic at hand and need to be wrangled back on topic. It would be natural to fill in the backstory and ask complex questions, and it would also be natural to have to dig in, restating some of those questions based on the answers you got back.” So, as you compose your first prompt in the conversation, be as clear as you can, including background, even noting what you do not want in the response, and be ready to post follow-up prompts to refine the output. One of my favorite generative AI applications is Perplexity.ai which automatically supplies three possible follow-up prompts that may give you the additional information you may be seeking.

Neuroflash, a leading European text and image generator, recommends that you craft your prompts to be precise and clear. Avoid long sentences with many sub points, instead, use short sentences that are easy to understand. Always try to ask specific questions that explain the context of the question.

At this early point in the advent of generative AI apps, many users simply enter the same very brief text they are accustomed to entering into Google Search, Bing, or other search engines. However, much of the power of generative AI comes in the ability of apps to customize the responses to the specific intended application of the results. The Decoder, a German-based digital publication on AI science, politics, and business, developed a list of ten prompt strategies. Among their top strategies are:

1. Define ChatGPT’s role: Linux terminal, philosopher, tea taster or bible translator – ChatGPT can play all these roles and many more.
2. Define target group and communication channel: To avoid tedious rewriting, either manually or through another prompt, you should give ChatGPT as much information as possible.
3. Chained prompting: A central aspect of getting ChatGPT to do what you want it to do is “chained prompting”.
4. Modify output: Even if you have followed all of the previous advice, you may want to prepare the same content for different channels, such as Linkedin, Twitter, or Facebook

Karrar Haider, writing in GeekFlare, reminds us of even more ways in which prompts can create unique, customized results including: “Use a witty tone while replying” or “Be comprehensive in your replies and give at least 3 examples.” Or you can ask the app to respond as if it were a professional in any specified field. Remember of course, the app is multilingual, fluent in many languages.  

The nominal limit for responses in ChatGPT is about 500 words per prompt. That may be too short for your purposes. As Fionna Agomuoh suggests in Digital Trends, “To remedy this issue, you can use simple prompts such as go on, keep going, or continue. For example, I inputted the query to write me a 1000-word essay on the history of Irish whiskey, and ChatGPT cut off mid-sentence at about 618 words. After inputting go on, ChatGPT continued generating my essay and ended at 1,043 words.”

Particularly for those of us in higher education, it is important to have citations in our work to provide further information and validate the information we have communicated. Some generative AI apps such as Perplexity.ai and You.com automatically include links to sources within the text of their responses, notably ChatGPT does not. That does not mean you cannot get citations from ChatGPT. ZDNet’s senior contributing editor David Gewirtz recommends that you follow the response to your prompt with the request to provide sources for the previous response. You may want to qualify that request by asking for URL sources in order that you can immediately dig into the material to verify facts or find additional information.

Daily rapid development of the technology continues. The newly-developed Auto-GPT has the ability to pursue a stated outcome by writing its own prompts and sequentially executing them. However, even this days-old technology needs a well-crafted, precisely-formulated outcome statement to begin its autonomous mission.

This article is far from a comprehensive discussion of generative AI prompts. Much more is to be learned by those who seek to become a “prompt engineer.” The rapidly-growing new career field is commanding high salaries without requiring high-tech degrees. The emphasis is on clear and comprehensive writing and problem solving. As written recently in Business Insider, “AI ‘prompt engineer’ jobs can pay up to $335,000 a year and don’t always require a background in tech.” Perhaps university English Department faculty members and Continuing Education Unit staff members should consider preparing themselves and their learners for these lucrative emerging positions!

This article was originally published in Inside Higher Ed’s Transforming Teaching & Learning blog.

In the flurry of launches of generative AI apps this year, there are some that I find are gems that don’t get enough attention. These are ones that I think colleagues in higher ed might use every day.

First, let me note that I have no investment in, have received no compensation from, or stand to gain from, the companies that produce or sell the apps mentioned in this article. These are applications that I have encountered in my readings for the UPCEA daily curation lists such as the Professional, Continuing and Online Update by UPCEA. I have tried them out, found them particularly useful, and put them in my “frequent-link box” so I can quickly access them throughout the day.

Let me begin with ChatGPT Plus. I found it well worth the $20 a month to get access to the GPT-4 version as well as the promise of front-of-the-line when times get busy. The new version, GPT-4, has the highest reasoning and conciseness ratings of the three versions that OpenAI supports: Default GPT-3.5, Legacy GPT-3.5 and GPT-4. If you use it once or twice a day, I think the plus version is well worthwhile. The time saved adds up to more than $20 in monthly wages.

I most often recommend this one as the first among alternatives to ChatGPT-4. Perplexity.AI is a top choice for colleagues in higher ed. I have found it to be a highly-reliable, quick-response generative AI chat. It is like ChatGPT with a few important differences. It is a search/chat interface that uses an OpenAI’s GPT 3.5 model to power its chat tool and search interface. Perplexity AI was founded by Aravind Srinivas, who worked as a research scientist at OpenAI. However, Perplexity AI is not a commercial product of OpenAI, but rather a demo inspired by OpenAI’s WebGPT. Perhaps the most valuable feature is that it automatically embeds citations within the responses! You have to ask for sources of information at the end of your searches done with ChatGPT. Having it presented up front, in response to every prompt, is useful, not only for those of us in higher ed, but everyone, to ensure the veracity and to uncover any potential bias in the response. Simply click on the URL-based citations, to check their relevance and accuracy.

Perplexity.AI also provides a short list of a few follow-ups in the form of possible prompts that it “thinks” you might want to consider to get a more complete response. Just click on one of those to get it added to your inquiry. Oh, and the initial responses are usually relatively concise – one or two paragraphs – that are ready for targeted expansion with follow-up prompts. There is no rambling, no padding, just the relevant information. In brief, it is a very quick-response tool that allows you to build a longer, more complex response – all with built-in citations.

Another of my favorites is ChatPDF. I love this app! Perhaps you are like me in that you are often viewing longer .pdf files. Government reports, commercial product promotional material, specification sheets, monthly/annual reports, financial reports, legal documents, training manuals, formal academic papers, book chapters, and so many more of our resources are held in .pdf or extended .docx format (which is easily converted to .pdf). This app responds to prompts such as summarizing, priority listing, a host of analysis approaches, and most all of the other GPT prompts to enable you to efficiently summarize and analyze your documents. The free version covers 120 pages, 10 megabytes, for three .pdf documents analyzed with 50 question prompts a day. The $5/month version covers 2,000 pages, 32 megabytes, for 50 documents and 1,000 question prompts a day. Your data is kept confidential in a secure cloud storage and can be deleted at any time.

I am a fan of Bing’s Image Creator. It is powered by DALL-E, which was revealed by OpenAI in a blog post in January 2021, and uses a version of GPT-3 modified to generate images. It is faster than some other image generators. As ChatGPT explained it to me (3/1/2023), “The difference between Bing’s image creator and DALL-E is that DALL-E is designed to create images that are more sophisticated and specific to the user’s input, whereas Bing’s image creator is more focused on quickly generating relevant images based on the user’s query. Overall, both Bing’s image creator and DALL-E use natural language processing to generate images, but the difference lies in the level of detail and sophistication of the generated images.”

The next generative AI favorite of mine is not really an app, rather it is an explanation of how to get GPT to create Excel spreadsheet formulas. This is a great utility for ChatGPT that fits right into the spreadsheet format that is favored by so many higher ed project directors and administrators. It also can write for Google Sheets which is popular in many shared environments.

Finally, here’s a look ahead into much more fully-integrated generative AI in business and productivity software. Eric Hal Schwartz writes in Voicebot.ai that Microsoft has released Microsoft 365 Copilot suite of models to 20 business customers that infuse generative AI into every Office application.

The samples are stunning! Schwartz explains: “The AI’s integration with Microsoft’s services allows it to turn raw data into Excel graphs, design and animate a PowerPoint presentation, and translate a meeting transcription into a long-term strategic plan for a company. The natural language prompts only require a vague description of the user’s goal and a connection to the relevant documents or meeting transcriptions to gather the data.” There are videos attached to the article that demonstrate the advanced capabilities that GPT performs using only very cryptic prompts. It is unclear if this will become a part of Microsoft 365 and if it will carry additional charges.

Those are my hidden generative gems, from the simple to the resplendent. Do you find any of these to hold possible promise for your work? Will they enable you to be more efficient and effective at your job? As AI expert Jaspreet Bindra is quoted as saying “ChatGPT will not take your jobs, someone who knows how to use it, will.”

This article was originally published in Inside Higher Ed’s Transforming Teaching + Learning blog.