When I discovered ChatGPT last winter, along with the rest of the human race, my initial reaction was one of bemused marvel. A month or two later, it occurred to me it might be gunning for my job. I suspect I’m not the only one who has felt wonder and revulsion, almost simultaneously, watching these programs spit out written evidence of seeming human intelligence at lightning speed.
Now, though? I’m looking for ways to make AI useful. No doubt you are too. After all, generative AI is garnering comparisons to the printing press . . . nuclear bombs . . . and the invention of fire. Who wouldn’t want to harness the power of fire if it’s going around? No one wants to be the last caveman pounding out strips of raw meat.
Hence this article, which will almost certainly look comically naive and off-target in a few years. That aside, I want to take a stab at answering the question: How can ChatGPT (and other generative AI programs) help you produce the fundraising copy you need?
Here are a few preliminary observations, three cautionary and three suggestive, to set you on the path towards harnessing the power of generative AI in your day-to-day fundraising operations.
First: Garbage in, garbage out.
ChatGPT produces things for you with blazing speed, but it’s very difficult to get it to give you what you’re actually looking for. In wrestling with it through numerous rounds of promptings and revisions, I start to feel like Arthur Dent in The Hitchhiker’s Guide to the Galaxy, spending untold hours talking to the Nutri-Matic beverage machine only for it to invariably produce “a liquid that was almost, but not quite, entirely unlike tea.”
This is probably the most important thing to understand about ChatGPT: The results it gives are only as good as the prompts you give it. Ask a lazy question, get a lazy response.
For example: In writing this article, I put in the following prompt: “Please write me a 500-word article about using generative AI programs like ChatGPT in fundraising writing.”
And here’s the opening paragraph of what it churned out, a little ditty called “Leveraging Generative AI Programs like ChatGPT in Fundraising Writing”:
The intersection of artificial intelligence (AI) and fundraising is a fascinating convergence of cutting-edge technology and the timeless art of inspiring generosity. As the demand for personalized, relevant, and engaging content increases in the fundraising sector, AI-driven tools, especially generative programs like ChatGPT, are taking center stage in addressing these needs.
Meh. It’s not awful (I like, for instance, the “convergence of cutting-edge technology and the timeless art of inspiring generosity.”) But for the most part, it’s boring . . . soulless . . . exactly the kind of dreck you’d expect a computer to spit out.
But look at the prompt I gave it. There’s precious little in those instructions besides “write me a piece about X.”
So I tried again, this time giving it guidelines on style, content, tone, and temperature (basically, how much or how little randomness and variability the algorithm will accept in producing new language). Here’s my prompt:
“Rewrite the article in the style of an Atlantic Monthly article, beginning with a striking anecdote. The revised article should integrate the five points into a more coherent flow, with several quotes from expert sources. The tone should be witty but concerned, and the temperature 0.7.”
And here’s what it gave me:
When a Machine Wrote an Appeal: Generative AI’s Foray into Fundraising
On a chilly morning in October, Susan, the director of a mid-sized nonprofit in Boston, received an appeal letter unlike any other. It was from a charity she’d supported for years, but this letter was uncannily perfect, tailored in a way that seemed to understand her past donations, her interests, even her passions. Two weeks later, at a charity event, she discovered the secret behind the letter: it was penned not by a human, but by an artificial intelligence program called ChatGPT.
Fundraising, the sacred dance of human connection, empathy, and generosity, is on the cusp of an AI revolution. From personalized outreach to crafting compelling narratives, programs like ChatGPT are reshaping how organizations communicate with their donor base.
“I’ve seen shifts in fundraising strategies over the decades, but AI-driven content? That’s an entirely new ballgame,” remarks Dr. Elizabeth Grant, a leading expert in nonprofit strategy at Harvard University. “It brings efficiency, but we need to ask, at what cost to the authenticity of our message?”
Pretty impressive, no? Still has that patina of soullessness, but it’s trending in the right direction. In my experience, the AI does its best work when you ask it to match styles. But the above example highlights a second problem with relying on generative AI.
Second: You cannot trust AI to provide you with reliable information.
Sad but true. In the above example, for instance, “Dr. Elizabeth Grant, a leading expert in nonprofit strategy at Harvard University,” was completely fabricated. Needless to say, there’s no evidence (on the internet, anyway) that this fictional person in fact discussed new ballgames or authenticity. ChatGPT made it up—as it did with the other two “expert quotes” requested in my prompt.
This is a problem. If we have no guarantee that generative AI programs won’t “hallucinate” our requested information, we gamble our accuracy and credibility every time we ask an AI program to come up with details for us.
It’s not just in generating new material, either. A few months back, I asked ChatGPT to summarize a long form essay I’d just read. Two of the four “takeaway points” it gave were fundamentally wrong—misrepresenting or inventing what was in the article.
Third: ChatGPT seems to suffer from short-term memory loss.
Maybe the problem is with me (if so, let me know in the comments!), but I cannot get ChatGPT to remember preceding steps or instructions, which makes complex operations difficult. When I tell it I’m going to teach it a style by feeding it different examples, it forgets after the first example what it’s doing and starts offering unwelcome analysis and pointers about the examples. When I give ask it to revise a document in one way, it forgets what the preceding requirements for the document were. In my experience, it struggles with complex writing in large part because it can’t remember anything you told it one step previously.
If you’re hoping ChatGPT can step in and handle all your writing needs, those are some hurdles you have to clear. It churns out boring, soulless prose. It makes stuff up. It can’t remember what you asked it a little while ago.
But did I mention it’s fast? Amazingly fast. That’s what an awful lot of its allure boils down to. It might give you subpar material, but it gives it to you instantly, with almost zero work on your part. It could be the holy grail of America’s obsession with low-hanging convenience.
Such snark sells generative AI short, however. This is a powerful tool. Here are three ways you can make it work for you:
1. Use it as an instructor.
Instead of relying on generative AI to catch a fish for you, ask it to teach you how to fish. It can generally crank out very good, detailed guidelines for how to write a specific kind of genre. It will provide you with templates and bulleted lists, putting you well on the way towards a solid first draft.
2. Use it for feedback.
There are certain things I would never trust ChatGPT on, such as making editorial adjustments. In one letter I asked it to evaluate and revise, almost every sentence it changed was changed for the worse. Its advice also skews milquetoast and establishment: it cautions you against things that might be edgy, controversial, or upsetting.
That said, it is tremendously useful when you ask it for a bulleted list of corrections or suggestions. It catches mistakes and makes good suggestions. It’s great at highlighting complex or confusing language. And if you give it specific terms to evaluate (“How could I make this email more donor-centric?”), it often offers very astute suggestions.
3. Give it discrete tasks.
As I mentioned in the first part of this article, if you ask ChatGPT (or any such generative AI program) to produce something long or complex, you’ll probably end up frustrated—in possession of a wealth of text that looks almost right but isn’t actually useful.
Your chances of success skyrocket when you assign short, discrete tasks. Ask it for ten teaser ideas for your mailing. Request that it outline the main objections to your argument or product pitch. Ask it to turn a complex paragraph into a simple bulleted list. Ask it to summarize and restate the main idea of your letter for the postscript. It can summarize, simplify, outline, rephrase, and much more. If you get a hang for using ChatGPT in the small things, it will almost certainly save you time and money.
Think about ChatGPT like a swift, indefatigable intern. You’re not going to trust the intern to handle highly sensitive material or produce distinctively excellent copy, but that doesn’t mean the intern isn’t useful. And this one doesn’t ever get confused, show up late, or eat all the Oreos.
“World’s fastest intern” might not sound like something on par with the invention of fire, but we’re only scratching the surface of what these generative AI programs can do. In a few years, when the robots are nuking the last human strongholds, this will all seem terribly quaint. Until then, good luck working ChatGPT into your fundraising writing. It can do remarkable things. Don’t forget to sound like a human, though.
The post Can a robot write a letter? appeared first on Philanthropy Daily.
Recent Comments