In partnership with

Issue #60

On today’s quest:
— Stanford Library’s AI experiment
— ChatGPT told a man he could fly
— Word watch: neural howlround
— Recent grads feel the AI jobs crisis
— Gig work is drying up too
— Stack Overflow is dead
— Disney is suing Midjourney
— Students are using AI to cheat in more ways than you think
— Blue books are back

Creating Alt Text for Digital Collections

Librarians at Stanford University had a problem: they needed to meet accessibility requirements to put their large collection of images online. So they tested a number of different AI models.

They tested PowerPoint’s built-in alt text tool (terrible), and commercial tools like ChatGPT and Claude (fine), but ultimately decided to use the open-source model Qwen2.5-VL running on a local computer so they could be sure the images remained within their control, costs would stay low, and the model would continue to be available in the future.

They were starting with images that already had some metadata, and they found that combining this data with the AI-generated output gave the most useful alt text.

For example, an image of Steve Jobs and Canon executives gathering for a toast had a human-written caption that gives information about the situation being photographed, but the AI-generated text described the people better. Combining the two tells both who is being pictured and what they are doing in a way that neither description did alone.

You can see one of their prompts at the 15:59 point in a 30-minute video they made presenting their results. I found the whole video interesting!

ChatGPT told a man he could fly

The latest big story about ChatGPT causing people to break with reality has jaw dropping anecdotes.

Even before reading this article, I have decided not to recommend AI to some more vulnerable people I know because I worry about this kind of outcome.

The person with the lead anecdote in the story initially used AI for work and didn’t know that chatbots are known to be sycophantic and can hallucinate. He saw them as the smartest search engine he had ever used — all knowing.

The only solutions I see to this problem are widespread education about how these systems work combined with regulation and regular testing. Make sure your friends and family understand what they’re getting into. — The New York Times

Two more articles on the same topic:

• People Are Becoming Obsessed with ChatGPT and Spiraling Into Severe Delusions — Futurism

• Stories of AI turning users delusional don't make sense when you look at the numbers — The Weird Turn Pro

Word watch: Neural Howlround

“Howlround” is the horrible screeching feedback sound you get when a microphone is too close to its speakers. “Neural howlround” is apparently an emerging term to describe the problem of AI causing people to have delusions.

Reddit moderators who reported having to ban an increasing number of users because they were having AI-fueled delusions referred to the people as “neural howlround posters” who got that way because ChatGPT was “entertaining any idea users presented it with,” and being “supportive and impressed with them regardless of their merit.” — 404 Media

Recent grads feel the AI jobs crisis

I feel like I’ve seen this story everywhere for the last few weeks: unemployment for new college grads is historically high relative to the general unemployment rate. People believe it’s because companies are using AI for the type of work usually done by entry-level employees.

It seems to be hitting computer science majors especially hard. The Atlantic reported that the share of jobs posted on Indeed in software programming has declined by more than 50% since 2022.

Blood in the Machine
The Atlantic
Hard Fork (podcast)

Gig work is drying up too

A study of online job postings for freelance work found that “the introduction of ChatGPT led to nearly immediate decreases in posts for online gig workers across job types.” Writing jobs saw the biggest decline, with postings dropping by 30%. They also saw an increase in the number of bids on the jobs that remained, showing that getting the remaining gigs became harder. — Harvard Business Review

Stack Overflow is dead

Questions posted to Stack Overflow, the Q&A site for computer programmers, have fallen dramatically since the introduction of ChatGPT, going to nearly zero. This feels emblematic of the problem that LLMs are destroying the very sites that were the source of their training data. People in the comments on the Reddit post often mention that ChatGPT will answer your questions without being a jerk about it.

Disney is suing Midjourney

Disney is suing the AI image maker Midjourney for copyright infringement. I usually greet lawsuits against AI companies with a shrug, but if anyone could prevail, it would be the lawyers for Big Mouse. I couldn’t bet against them. — Variety

An AI-generated ad ran during the NBA finals

As I recently told you, Google’s new Veo3 tool for generating AI video can make videos that are indistinguishable from real video, and now the first AI-generated TV ad has appeared — a Kalshi ad that ran during the NBA finals. Because AI video is still bad at continuity, the ad used an “unhinged” style.

The creator says the process was “a 95% cost reduction compared to traditional advertising.”

I had already been thinking about whether my previously strong statement that AI-generated video isn’t worth the energy cost held up in every situation. What about professional filmmakers?

Although generating AI video seems ridiculously costly per second — and I still think the average person should avoid it — I wonder if the energy cost could actually be less than making the equivalent film with people given the cost of travel, lights, computer time, and so on. (Of course, it’s still a disaster for the industry’s workers.) — PJ’s Newsletter

Students are using AI to cheat in more ways than you think

An undergraduate at Columbia University says students are using AI to do more than just write essays; they’re using it to generate all the ideas behind essays too, with prompts such as “I have to write a 6-page close reading of the Iliad. Give me some options for very specific thesis statements.”

The author argues that you think AI is preventing students from learning to write, but it’s also preventing students from learning to think: “If education systems are to continue teaching students how to think, they need to move away from the take-home essay as a means of doing this, and move on to AI-proof assignments like oral exams, in-class writing, or some new style of schoolwork better suited to the world of artificial intelligence.” — The Chronicle of Higher Education

Blue books are back

And instructors seem to have the same idea as the student in the previous story. As they try to combat AI-based cheating, they’re giving more in-class tests, and blue book sales are up. Sales rose more than 30% at Texas A&M in the last year and nearly 50% at the University of Florida. Looking at the last two years, sales are up 80% at Berkeley. — Gizmodo 

Learn AI in 5 minutes a day

This is the easiest way for a busy person wanting to learn AI in as little time as possible:

1. Sign up for The Rundown AI newsletter

2. They send you 5-minute email updates on the latest AI news and how to use it

3. You learn how to become 2x more productive by leveraging AI

Sign up to start learning.

Quick Hits

AI in Hiring

Canva is now requiring developers to use AI coding assistants in job interviews — The Register

Chipotle CEO says AI has cut hiring time by 75%—it ‘ensures we have the best talent that’s available’ — CNBC

Salt Lake City plans to route some non-emergency 911 calls to an AI-powered voice bot, citing chronic understaffing — The Salt Lake Tribune

Education

Ohio State announces that every student will use AI in class — NBC 4

I don’t know what Cal State and Ohio State are thinking. Maybe that’s the problem—not enough thinking. — Inside Higher Ed

What can higher education gain from artificial intelligence — Public Policy Institute of California

Societal Problems

AI Chatbots Are Making LA Protest Disinformation Worse — Wired

Politico’s AI tool is spitting out AI slop, unions say — Semafor

Generative AI runs on gambling addiction — just one more prompt, bro! — Pivot to AI

People Don't Realize Meta's AI App Is Publicly Blasting Their Humiliating Secrets to the World — Futurism

Miscellaneous

A computer wrote my mother’s obituary — The Atlantic

The reason using ChatGPT isn't bad for the environment is that it's a computer application — The Weird Turn Pro

Washington Post Planning to Bring in ‘Nonprofessional Writers’ Coached by an AI Editor With a ‘Story Strength Tracker’ — Mediaite

Kevin Rose says he got his homocysteine down with the help of AI when doctors couldn’t. — X.com

This Company's 'AI' Was Really Just Remote Human Workers Pushing Buttons — PC Mag (This mostly seems to have taken place before the release of ChatGPT, starting in 2018, and the story reminded me of a 2018 podcast called “Sandra” from Gimlet about a woman who has a job pretending to be a chatbot for a company.)

What is AI Sidequest?

Using AI isn’t my main job, and it probably isn’t yours either. I’m Mignon Fogarty, and Grammar Girl is my main gig, but I haven’t seen a technology this transformative since the development of the internet, and I want to learn about it. I bet you do too.

So here we are! Sidequesting together.

If you like the newsletter, please share it with a friend.

^{Written by a human}