How journalists are using AI

Issue #59

On Today’s quest:
— How journalists are using AI
— Simple ways to edit faster
— Vibe coding
— Corpora versus LLMs
— AI voices are now more expressive
— Quick hits

How journalists are using AI

The Columbia Journalism Review recently documented how journalists are using AI. All the profiled journalists opposed using AI to do their primary writing, but they cited many other uses — transcription being the most common.

One journalist noted that “roughly 10 percent of 2024 Pulitzer Prize finalists used AI in some capacity.”

Other AI uses the journalists mentioned included:

• Getting information out of large data sets, both text and images

• Researching people and companies they are going to interview

• Editing short-form videos and podcasts

• Translation (such as getting the gist of foreign YouTube videos)

• Getting summaries of research papers

• Getting headline suggestions

• Getting various kinds of feedback on drafts

• Generating illustrations for stories (Midjourney)

• Organizing files for large projects (NotebookLM)

The concerns the journalists mentioned included:

• Water and energy use

• The environmental problems disproportionally affect minorities

• The models were trained on the work of journalists without consent

• AI slop images makes it harder to do open-source intelligence investigations

• AI slop is muddying search results

• The flood of fake images is making it harder for people to know what is real

The article is definitely worth reading to see the details .

Through the CJR article I also found a free open-source AI course created by two of the profiled journalists.  The purpose is to “Learn how journalists use large-language models to organize and analyze massive datasets.”²

Simple ways to edit faster

Marcella Fecteau Weiner had a great blog post last week describing four simple ways she routinely uses AI to speed up her editing:

1. Writing more concisely

2. Reformatting bulleted lists

3. Writing Word macros

4. Drafting an email from notes

I have used AI to clean up notes too. I have a bad habit of emailing myself thoughts in a long thread as I’m preparing for interviews because I’m often on different computers, which means I end up with a mess that needs to be organized in the end. An LLM can get all the text cleaned up and into one document for me to review in seconds, saving me a lot of copying-and-pasting time.

Check out Marcella’s post to see her examples and prompts

Vibe coding

Vibe coding tools use AI to let you quickly create things that would normally require coding. Back in April, I used Gemini 2.5 to vibe code a version of my old iOS game, Grammar Pop, and it blew my mind. You can see it in action here.

It’s not beautiful, and it doesn’t have the 14,000 sentences that were included in my game, but it works, and it took 5 minutes — maybe less — and I was shook. It took me at least 4 months (and many tears) to get to this level of functionality when I was coding the game myself.¹

I wouldn’t release anything into the wild if I didn’t understand how it worked, but something like this could be amazing for rapid prototyping, and I’ve heard many stories of people quickly making apps for themselves to handle specific tasks.

Next week, I’m taking a $5 vibe coding workshop put on by Adam Hyde, who is starting Vibe Coding Workshops. He’s just starting out and is currently offering the workshop for this nominal price to test it on some people. I don’t know if it will be useful, but I’m looking forward to learning a bit more. The price goes up June 21.

Also: Why vibe coding your own apps is so amazing—and exasperating — Fast Company

LLMs Versus Corpus data

Corpora are huge collections of text that linguists often use to study language, and linguist Mark Davies has created many of the most widely use corpora available, which live at his english-corpora.org website.

Davies recently released a video comparing using LLMs to using corpora for different types of language research. For example, when he asked LLMs to estimate how common words or phrases are in general or in different dialects of English, the results matched data from corpora quite well. They were less good at generating lists of words from scratch, and ChatGPT did a better job than Gemini.

He especially liked that LLMs gave explanations for why the words collocated or were used differently in different countries. He thinks researchers should probably stick with corpora, but he can see how language learners might benefit from using LLMs. For both groups, he thinks the best scenario is probably to use the tools in combination. (I really enjoyed the whole video, but if you’re in a hurry, the conclusions start around 23 minutes in.) h/t Lynne Murphy

AI voices are now more expressive

Eleven Labs has a new feature that lets you direct an AI voice to have different emotions (e.g., outraged and whisper) and to add a few sound effects (e.g., applause and gulps).

See it in action in the video below. I like how it whispers! Capitalizing words makes them have more emphasis, and ellipses are supposed to create pauses (but the ellipses in my text didn’t seem to work — it’s still an experimental feature).

My example is just one try. Here’s another example from Ethan Mollick who spent more time trying different things to get it to work better. Also, you can see all the available tags at the Eleven Labs site.

Quick Hits

5 AI bots took our tough reading test. One was smartest — and it wasn’t ChatGPT — Washington Post

At Secret Math Meeting, Researchers Struggle to Outsmart AI — Scientific American

AI firms say they can’t respect copyright. These researchers tried — Washington Post

What I learned serving on my university’s AI committee — Chronicle of Higher Education

What AI can’t steal from you (by Jane Friedman) — The Writing Platform

Pulitzer-winning cartoonists are experimenting with AI image generators — Neiman Lab

OpenAI is retaining all ChatGPT logs “indefinitely.” Here’s who’s affected — Ars Technica

"Learn to Code" Backfires Spectacularly as Comp-Sci Majors Suddenly Have Sky-High Unemployment “Companies are cutting engineering budgets by 40 percent while CS enrollment hits record highs.” — Futurism

Welcome to campus. Here’s your ChatGPT “OpenAI, the maker of ChatGPT, has a plan to overhaul college education — by embedding its artificial intelligence tools in every facet of campus life.” — New York Times

AI Exhibits Racial Bias in Mortgage Underwriting Decisions “Researchers found that LLMs consistently recommended denying more loans and charging higher interest rates to Black applicants compared to otherwise identical white applicants.”— Lehigh University

PODCAST: I enjoyed the segment on the Hard Fork podcast this week about chefs using AI. Listen to the section from 47 minutes to the one-hour mark. (The link will start you there on YouTube.)

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

1. I taught myself to code to make Grammar Pop, so it would have taken a real developer a lot less time, but that’s the beauty of vibe coding—it lets noncoders make things easily.

2. I haven’t reviewed the course yet, so I can’t tell you how useful it will be. Also I was extremely annoyed that the article didn’t include the link, so I came down here to the footnotes to complain. I hate articles that mention resources and don’t link to them. I searched and searched and couldn’t find it, but ChatGPT found it in one prompt.

Written by a human