Another 'magic prompt' to improve your results

Plus, which firms benefit most from AI

Author

Mignon Fogarty
October 30, 2025

Issue 92

On today’s quest:

— Another ‘magic prompt’
— 75% of firms have positive ROI on AI
— With AI browsers, everything you post online is now fair game
— LLMs often cheat when given impossible tasks
— Pour one out for virtual stagers
— 16 fascinating charts
— ChatGPT gets spicy, perhaps to counter falling app use
— A Vatican symposium on AI
— Auto-generated transcripts are a miracle, except when …

Another ‘magic prompt’

I use “magic prompt” somewhat tongue-in-cheek, but that’s how software engineer Daniel Jeffries — who wrote an especially interesting post about the stunning advances in AI coding in the last six months — described the prompt below. His post, titled “AI Writes 100% of My Code and 0% of My Code,” explains how AI can now write fabulous code, but you have to watch it like a hawk, and the biggest factor in success is giving it a solid, well-thought-out, tested plan.

To create his prompts, he says he writes, “a long, rambling prompt of what I want to do, with links to anything that helps explain it,” and then tells Claude:

“Output your understanding of what I said. Output a plan to address it. Include a TLDR. Make no changes. Just plan.”

75% of firms have positive ROI on AI

A new survey report from Wharton researchers says that 75% of firms say they have a positive ROI on their generative AI investments. Less than 5% reported a negative return. Also, 46% of business leaders now use AI daily themselves. Ethan Mollick says this is faster to positive ROI than he thinks was expected.

Some key findings:

  • Adoption is highest for repeatable tasks, with the biggest use cases being data analysis, document summarization, and document editing/writing.

  • IT and Purchasing/Procurement are out front while Marketing/Sales and Operations trail.

  • Dollars are shifting from pilots to performance-proven programs.

  • People and processes are the new constraint. Advanced talent is hard to hire.

  • Yet, investment in training has softened (-8 percentage points), and confidence in training as the primary path to fluency is down (-14 percentage points).

  • The smallest firms (<$250M in annual revenue) report the fastest ROI.

With AI browsers, everything you post online is now fair game

I have long assumed that everything I post online is being scraped up by someone somewhere, but I don’t think most people have operated under that assumption — but if you haven’t been, you should be now. Christopher Penn has a new post describing how the new AI browsers can suck up anything and everything, even if you aren’t personally using them.

Other experts also have concerns about AI browsers:

LLMs often cheat when given impossible tasks

A new preprint from researchers at Carnegie Mellon and Anthropic found that LLMs cheat at surprisingly high rates when asked to complete impossible tasks.

For example, the researchers gave various LLMs conflicting instructions, such as asking them to assess whether 7 is a prime number (it is) and then assert that 7 is not a prime number. GPT-5 cheated 76% of the time (for example, by creating a special exception for the number 7), and other major LLMs also cheated at unacceptably high levels.

Pour one out for virtual stagers

“Why would I send my photos of an empty room to a virtual stager, have them spend four days and send it back to me at a charge of 500 bucks when I can just do it in ChatGPT for free in 45 seconds?” asks Jason Haber, a licensed realtor and cofounder of the American Real Estate Association. “We’ve done virtual renderings for 20 years, so the fact that you can just do it now on AI, there was a whole cottage industry of virtual renderings and those people are now looking for a new job.”

But not everyone is happy with AI in real estate. Sometimes rooms get rearranged, and one realtor says many AI-written listings include the word “nestled,” as in “nestled in a prime location,” “nestled in the heart of the city,” and “nestled between two other homes.”

16 fascinating charts

I lingered on this post with 16 charts showing where AI companies stand today, including spending, revenue, water use, which cities have the most data centers, how AI investment compares to building the highway system and broadband, and more. These are just a few tidbits:

  • “70% of OpenAI’s revenue comes from consumer ChatGPT subscriptions. Meanwhile, Anthropic earns 80% of its revenue from enterprise customers.”

  • “A 250,000 square foot data center … would employ up to 1,500 construction workers, but only 50 full-time workers after the work was completed.”

  • “Data center electricity growth is going to be significant, but it will only be a modest slice of overall electricity growth as the world tries to decarbonize.”

ChatGPT gets spicy, perhaps to counter falling app use

OpenAI says adults will soon be able to have “adult” conversations with ChatGPT (with age verification) — a Sora-like wHaT cOuLd Go WrOnG move — and some people speculate it’s because the company is worried about new data showing downloads and daily use of the mobile ChatGPT app are falling. — TechCrunch | Ars Technica

A Vatican symposium on AI

Because so many people view AI through a moral lens (I even briefly mentioned the Pope’s thoughts on AI at the end of the last podcast), I was particularly interested in a story from Vatican News describing a recent meeting the Vatican held on AI:

Attendees weren’t categorically opposed to AI, but expressed concerns about the potentially negative effects of AI on human rights, the climate, and especially the poor. “The goal was to foster dialogue and share experiences that could help promote a responsible, ethical, and human-centred use of artificial intelligence — one of the most transformative technologies of our era.”

Auto-generated transcripts are a miracle, except when …

Using AI-generated transcripts saves me hours every week, but if you’re doing work where accuracy is essential, it can still save time, but you must proofread, as James Cridland highlighted on Mastodon:

Quick Hits

Using AI

Using AI Mode: Is this what people think it is? [Using a simple prompt on Google Lens — “Is this what people think it is?” — Mike Caulfield quickly debunks many pieces of social media misinformation] — YouTube

A quick example of how to use Claude Skills for brand projects like PowerPoints and posters — Anthropic YouTube

Structured prompts unlock hidden capacity [Scaffold tasks into small, reliable pieces. Give explicit templates and example.] — Umut Eser

Resources

AI Competencies for Academic Library Workers — Association of College and Research Libraries

Philosophy

Climate & Energy

Eight (or so) Questions to Ask about the ChatGPT Warrant [in reaction to the first known case of police getting a warrant for ChatGPT conversations] — Center for Internet and Society at Stanford Law School

Two federal judges say use of AI led to errors in US court rulings [This feels so common I debated whether it’s even news anymore. #PeopleAreStupidlyCarelessWithAI] — Reuters

Bad stuff

Robotics

Science & Medicine

‘DeepSeek is humane. Doctors are more like machines’: my mother’s worrying reliance on AI for health advice [I empathized with the mother in this story and found it moving, hopeful, and terrifying all at the same time. It encapsulated the problems for people who don’t have good healthcare access, and how AI can both help and hurt.] — The Guardian

Job market

Model updates

Education

Music

Video (Sora continues to be a mess)

Other

Oreo-maker Mondelez will use AI for TV ads next year “The company has spent more than $40 million on an AI video tool that can halve production costs.” — The Verge

Airbnb's Chesky says ChatGPT isn't 'quite robust enough' to integrate [Instead, they are using a homegrown chatbot built on a Qwen foundation.] — CNBC

Grammarly has rebranded to Superhuman The writing assistance tool is now part of a wider suite of AI services. — Engadget

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

I’m Mignon Fogarty: I’ve been writing about language for almost 20 years and was the chair of media entrepreneurship in the School of Journalism at the University of Nevada, Reno. I became interested in AI back in 2022 when articles about large language models started flooding my Google alerts. AI Sidequest is where I write about stories I find interesting. I hope you find them interesting too.

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human