Issue 84

On today’s quest:

— Judge postpones Anthropic settlement
— Tip: massaging your online bios for AI
— AI reduces junior-level hiring
— For editors: learn about AI
— How to opt out of Anthropic data sharing
— Removing private data from LLMs
— Weird AI: Oh … it speaks every language
— Google says the open web is in decline
— 120+ vibe coded tools
— Follow-up: compare documents

Judge postpones Anthropic settlement

On Monday, Judge William Alsup threw a wrench into the class-action Anthropic copyright case, asking for more details before he will decide whether to approve the $1.5 billion settlement. He wants:

• A firm list of all the books that will be eligible for a payout.

• A clear process for notifying class members.

• A limit on the amount of the proceeds that can go to lawyers, citing concerns about how many lawyers had been added to the process after the settlement was announced, calling them “hangers on.”

The next hearing will be September 25th, at which time Alsup said, “We’ll see if I can hold my nose and approve it,” according to the Associated Press.

If you want more information about what this all means for authors, check out my new experimental AI Sidequest podcast! We did a “breaking news” show Monday night. It’s on both audio and YouTube:

Tip: Massaging your online bios for AI

Jess Zafarris had a great tip on LinkedIn about checking what AI knows about you. In short, type “Who is [your name]?” into a chatbot (the equivalent of old-school Googling yourself), and see what it says. If you don’t like the focus, try changing your bio on the places it is drawing the information from if you can.

She said she saw changes within a couple of days.

AI reduces junior-level hiring

A new large-scale study found that, starting in Q1 of 2023, firms that had adopted AI hired fewer junior-level employees than those that hadn’t adopted AI.

The biggest drops were seen in the “wholesale and retail trade,” which likely includes jobs doing work such as customer service, data entry, and creating basic marketing content. People graduating from mid-tier colleges were more affected than people graduating from elite or low-tier colleges.

A couple of caveats are that this paper is a preprint, so it hasn’t been peer reviewed yet, and only 3.7% of the companies in the study were identified as having adopted AI, so although the researchers looked at a lot of companies overall and the results were statistically significant, AI adoption was still a small number.

Interestingly, the researchers used AI to scan the job listings of companies for AI-related keywords to identify companies that had adopted AI.

For editors: Learn about AI

Erin Servais of AI for Editors recently hosted a private discussion with great examples of how editors can use AI in their fiction and nonfiction work, and now the replay is available on YouTube. Here’s the breakdown:

• 9:00 Marcella Fecteau Weiner. Nonfiction editing: Using AI to make sure lists are parallel and to format citations.

• 20:40 Kristen Tate. Fiction editing: Using AI for consistency checking.

• 35:00 Erin Servais: Marketing: Using AI to build customer/reader personas.

• 46:00 Questions.

Kristen Tate also recommended WisprFlow, an AI dictation tool I have recommended before. If you’d like a 30-day free trial, my affiliate link is https://wisprflow.ai/r?WISPR11967. I love that you can use it in a less-than-perfectly-quiet room and that you don’t have to speak the punctuation marks — it just knows where to put them.

How to opt out of Anthropic data sharing

In the past, Anthropic didn’t use your chats to train new models. But that has changed. Starting September 28, you have to opt out or your chats will be used for training.

According to TechCrunch, “the new policies apply to Claude Free, Pro, and Max users, including those using Claude Code. Business customers using Claude Gov, Claude for Work, Claude for Education, or API access will be unaffected.” 

The settings aren’t easy to find. On the desktop, click Settings → Privacy and then click the black Review button. 

Then, toggle the “You can help improve Claude” button off, so that it’s gray. (It was still on in my settings below when I took the screenshot.)

Removing private data from LLMs

Researchers at UC Riverside have come up with a way to make LLMs “forget” information, which would let companies remove private and copyrighted data from models after they’ve been trained — something that couldn’t be done before. The researchers said the technology could “empower people to demand the removal of personal or copyrighted content from AI systems.”

Beehiiv: How I make this newsletter

I make this newsletter with Beehiiv, and I’ve been super happy with the CMS and features. If you’ve been thinking of ditching your current newsletter platform (or starting a new newsletter), I heartily recommend it.

Use my affiliate link to get a 30-day free trial and 20% off your first three months. I’m happy to answer any questions you have about the platform too!

Weird AI: Oh … it speaks every language

Sal Kahn of the Kahn Academy was a guest on the Decoder podcast and told a funny story about OpenAI not understanding what ChatGPT-3.5 could do (podcast link):

Google says the open web is in decline

In a court filing to defend itself against claims it is an advertising monopoly, Google said the open web is in rapid decline. Interestingly, this contradicts statements it has made to website owners about how Google AI Overviews are not crushing website search traffic. (As a website owner whose traffic has declined, I’m biased and feel quite prickly about this.)

120+ vibe coded tools

Simon Willison is going a little wild vibe coding small single-use tools, which he calls “an experiment in prompt-driven development with very low stakes.” Many of his 120+ tools are meant for coders, but you’ll also find a simple word counter, time-zone converter, reading time calculator, phonetic alphabet converter, various PDF processing tools, and a few useful image resizing tools.

Follow-up: Compare documents

In the last newsletter, I told you about using ChatGPT to compare two Google Docs to find changes between them.

I’m always happy to use a non-AI solution when it’s as good as what I can get with AI, and Phil Simon told me about a site called Diffchecker. It’s built for code, but also works on text, and does the job quickly and at about the same level as ChatGPT.

Quick Hits

• ChatGPT on GPT‑4o Mini Found Giving Higher‑Risk Suicide Responses — Konsyse

Using AI

• How to use “sorting statements” to reduce hallucinations — Mike Caulfield

Climate

• Mapping the AI & environment debate (A good big-picture piece outlining this author’s view on AI and climate)— The Weird Turn Pro

• The water question is so complicated it makes my head hurt. Some estimates include the amount of water that evaporates from dams. — Construction Physics

Legal

• AI Chatbots Are ‘Clear Danger’ to Kids, Australian Watchdog Says. New law says sites, including AI chatbots, must use age-checking technology to stop children under 16 from accessing unsafe material — Bloomberg

I’m laughing

• I hate my AI friend. Two reporters hate their AI wearable, called Friend — one because it’s socially unacceptable to record everyone around you, the other because the device was a real jerk. — Wired

Job market

• AI threatens entry-level marketing jobs — and the future talent pipeline — MarTech

• GenAI and the future of government work (PDF) — IBM

Model updates

• Google rolling out Flashcards, Quizzes, and redesigned Reports to NotebookLM — 9to5Google

• Gemini app finally expands to audio files — The Verge

Education

• Bluebooks: The only real solution to the AI cheating crisis — The New York Times

• I got an AI to impersonate me and teach me my own course — here’s what I learned about the future of education — The Conversation

The business of AI

• AI progress may slow because data centers and chip factories are operating at capacity, and it takes a long time to build new ones — Epoch AI

Other

• We’re Walling Off The Open Internet To Stop AI—And It May End Up Breaking Everything Else — Techdirt

• Will Artificial Intelligence Do More Harm than Good to the U.S. Economy? “AI is contributing to U.S. growth in at least two major ways: massive capital expenditures and wealth creation. At the same time, it is creating what could become a meaningful drag—job cuts.” — Council on Foreign Relations

• Why 95% of AI Commentary Fails. In which I actually read the paper everyone is talking about. — Gradient Ascendant

• 5,000 Podcasts. 3,000 Episodes a Week. $1 Cost Per Episode — Behind an AI Start Up’s Plan — Hollywood Reporter

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 (except the examples below, obviously)