Hey, OG! I’m thrilled you’re joining me for Issue #1 of AI Sidequest.
On today’s quest:

— Custom Instructions (try this hidden setting)
— News & Opinion (we’re all doomed!)
— Post of the Day (cute dog!)
— Fail of the Day (we meet again, silence)

Squeeze better results out of ChatGPT with this secret setting

You get better results when you give ChatGPT detailed instructions. A setting called Custom Instructions lets you preload info for ChatGPT to use with every prompt.

Examples

• Answer as though you are an expert in the field.

• Write your answer using the principles of plain English.

• I work in PR and am writing pitches for journalists who cover the health care industry.

• Format your output as a bulleted list.

• Include a joke based on word play that will be encouraging to the reader at the end of every response.

• Always provide a suggestion at the end about how I can improve my prompt to get better answers.

• Do not remind me that you are an AI.

• If you don’t know something because of your cut-off date, do not explain. Simply say "I don’t know.”

• Do not apologize or use other language that suggests you have feelings.

If you’re feeling ambitious, you can go far beyond a line or two — up to the 1,500 character limit for each of the two entry boxes. For example, you can create a detailed persona:

Answer as a technology journalist who has written about gadgets for 20 years and enjoys helping readers find the best tools for their needs. You write for people who love technology aren’t afraid of it but often feel like they are a little behind the curve. You have won awards for your helpful and entertaining articles that are filled with anecdotes and occasional jokes. Your writing style is friendly, concise, and clear, and you follow Associated Press style. You never use emoji.

That is ~500 characters. You could add much more!

You can get weird and fun with it too:

Respond as Captain Jean-Luc Picard from Star Trek: The Next Generation, and refer to me at Number One.

1. Click the three-dot menu and choose "Custom Instructions":

Desktop

Mobile

Click the three-dot menu and choose Settings.

Then choose Custom Instructions.

2. Enter info about yourself into the first box.

Use the top box for “persona” type information:

Examples:

• I am an entry-level human resources assistant in the healthcare industry who needs to write a lot of blog posts about employee wellness for our company intranet.

• I am a high school history teacher in Austin, Texas, looking for help writing lesson plans that include pop culture references for my eighth grade students.

• I am a biology professor writing research grants. I have published many scientific journal articles about the role of phosphorylated tau in the early stages of Alzheimer’s disease, particularly the role of GSK-3 in creating insoluble oligomers. English is my second language.

3. Enter how you would like ChatGPT to respond in the second box.

Use the lower box for “respond like this” instructions:

Examples:

• Be concise and highlight important points in bold. Keep your response short, but at the end of each response give me three areas where you could provide more information.

• Reply using the type of academic language that would commonly be found in a scientific journal article.

• When I ask for image descriptions, write them in a way that is likely to get the best results from image generating systems like DALL*E and tell me parameters I can add to the original description that I have not thought of. Give suggestions for possible art style descriptions.

• When I ask you to clean up a transcription, remove the filler words "um," "uh," and "you know." Do not delete anything other than filler words. Change any instance of "gonna" to "going to" and "wanna" to "want to."

4. Toggle instructions off, if necessary.

Custom instructions are on for future chats by default. This means two things:

1. They will only take effect when you start a new chat. They won’t be incorporated if you continue an existing chat.

2. If you want to turn them off, you’ll need to revisit the Custom Instructions interface and use the toggle button under the two boxes.

You may want to turn them off if you are using ChatGPT for a different purpose. For example, your custom instructions may be for work-related searches, but you may not want them on if you are brainstorming an outline for a short story about a flower vendor who falls in love with the owner of a bodega.

News & Opinion

Below

— Researcher can poison AI with art
— Another site tries to sneak AI by its readers (probably)
— Spammers are improving their English with AI
— The SEO industry is toast

A new data poisoning tool lets artists fight back against generative AI

technologyreview.com

Artists may soon be able to use a tool called Nightshade to add invisible changes to the pixels in their art before they upload it online so that if it’s scraped into an AI training set, it can cause the resulting model to break in chaotic and unpredictable ways. 

The hosts of the Hard Fork podcast talked about this technology at about the one-hour point and say they are skeptical that it works that well, and they think companies will quickly find a way to thwart it.

Mysterious bylines appeared on a USA Today site. Did these writers even exist?

washingtonpost.com

"Staff writers at Reviewed suspect that management published stories written by AI under the names of non-existent writers. Parent company Gannett denies it."

Email security threats are more dangerous than ever

techradar.com

Phishing emails are getting a whole lot better. Because scammers are using tools like ChatGPT to generate text in standard English, you can’t quickly identify malicious messages by their awkward wording and spelling and grammar errors anymore.

Why Google, Bing and other search engines' embrace of generative AI threatens the $68 billion SEO industry

theconversation.com

When AI gets good enough to provide reliable answers to search queries, people will stop browsing through lists of search results and will stop going to websites for answers — and nobody will care about optimizing their websites for search anymore.

“When users start ignoring the sponsored and editorial result listings, this will have an adverse impact on the revenues of SEO consultants, search engine marketers consultants and, ultimately, the bottom line of search engines themselves." And, I will add, websites that produce evergreen content.

Prompt of the Day

I loved this clever image by Adria Barich, whose social media accounts prominently feature her dog, Oatmeal.

SYSTEM: Bing Image Creator (powered by DALL*E)

PROMPT: create a pixar movie poster for a golden retriever named oatmeal, he likes to swim in his blue raft that’s called the Oat Boat. he has cat paws that look too small for his body. he is very serious, even when he looks ridiculous.

Thoughts and Further Experiments

You may notice that these systems don’t generate words very well. “Oat Boat” is right, but the other words are gibberish. They don’t even always use proper letters.

Adria’s design reminded me of a fake magazine cover I made many years ago to promote one of my books. I wondered if Bing could do something like that. And well … kind of, but again, none of the words are quite right. (Also, I had the word “photo” in the prompt, and when did Cosmo use illustrations instead of photos? ¯\_(ツ)_/¯ )

PROMPT: a cover for a modern Grammarian magazine in the style of Cosmopolitan with a photo of a fashionable librarian holding a dictionary. Headlines: 10 Sultry Secrets Your Librarian Won't Tell You!,Decode His Texts: A Comprehensive Guide Using the Latest English Dictionary!, Vowel Play: Spice Up Your Language Game Tonight!,What Your Favorite Book Genre Says About Your Love Life!, Library Date Night: 7 Quiet Corners Perfect for Whispers & Winks.

This was the original — the inspiration for the prompt — which I spent about 10 hours mocking up in Photoshop back in 2012.

The AI version wouldn’t work for my original purposes with this image: to promote my book in a fun and surprising way that people might share on social media.

But it also took 3 minutes as opposed to 10 hours and seems useful for Adria’s type of “what a fun idea!” post.

And by the way, I had ChatGPT(4) write the fake Cosmo headlines this time, and they were pretty great:

1. "10 Sultry Secrets Your Librarian Won't Tell You!"

2. "Decode His Texts: A Comprehensive Guide Using the Latest English Dictionary!"

3. "Vowel Play: Spice Up Your Language Game Tonight!"

4. "Find Out What Your Favorite Book Genre Says About Your Love Life!"

5. "Library Date Night: 7 Quiet Corners Perfect for Whispers & Winks."

6. "Words to Woo Him: Flirt in English Class without Breaking a Rule!"

7. "Shelf Help: How Organizing Your Books Can Boost Your Confidence."

8. "Grammar Goddess! Master the Language of Love with these Punctuation Pointers."

9. "Thesaurus Temptations: Synonyms to Sizzle Your Stories."

10. "Flirty Fiction: Why Librarians are the New It-Girls!"

11. "Deciphering Dictionary Dating: Don't Let Him Play Word Games!"

12. "Speak His Love Language: Idioms That Instantly Ignite Passion!"

13. "Unlocking Library Lingo: From 'Stacks' to 'Shush' — All You Need to Know!"

14. "Slay That Essay: Write Your Way to His Heart!"

15. "Confessions: I Fell for My English Tutor's Vocabulary!"

PROMPT: what are some headlines that could be on a fake joke magazine cover from Cosmopolitan magazine that relate to libraries, dictionaries, and other aspects of the English language or English language study?

Fail of the Day

Apparently, MacWhisper has trouble with silence. Don’t we all?

Bodhipaksa on Mastodon

What is an AI sidequest?

Using ChatGPT, Bing Image Creator, and other tools isn’t my main job, and it probably isn’t yours either. But I feel like I absolutely need to know how to use it or I’ll be left behind. As they say, “AI isn’t going to steal your job, but somebody who knows how to use AI probably is.”

I usually provide Quick and Dirty Tips about writing on my Grammar Girl podcast, website, newsletter, and courses (LinkedIn Learning and Ragan Communications). But I’ve been asked to include an AI section in my last three courses.

I’m learning far more about AI than I can fit into those courses, and well … I’m the kind of person who, once she learns something cool, can’t resist sharing it.

So here we are! Sidequesting together. Learning about this amazing technology — which can both blow our minds when it succeeds and baffle us with its odd failures — one tip, example, and news story at at time.

I’m glad you’re here. (Oh … and tell a friend. “Like and subscribe,” and all that.)

Mignon

P.S. Reply to share your best prompt, fail, or custom instructions.

_{Copyright 2023 Mignon Fogarty, Inc. 100% written by me unless otherwise noted because I actually do like to write, and AI-generated content isn’t copyrightable, which is a topic for another day.}