Issue #25. With great excitement and appropriate caution

On today’s quest:

— Teachers get an AI sandwich
— Canva jockeys get better tools for making up quotations
— Artists get a tool for fighting back

Tip: Teaching with AI

Today’s tagline, “with great excitement and appropriate caution,” is how Chris Reykdal, superintendent of Washington State’s public schools, says they are approaching the use of AI in K-12 schools. “AI is here, and slowing down isn’t an option.”

The 19-page guidance document (pdf) encourages a sandwich approach, with AI in the middle: “Start with human inquiry, see what AI produces, and always close with human reflection, human edits, and human understanding of what was produced. … AI should aid in (not replace) decision-making, creativity, learning, development, growth, and productivity.”

The report covers both high-minded goals (help students develop AI literacy, ensure that AI use promotes critical thinking) and practical goals (streamline administrative functions).

The report imagines schools using AI for “differentiation and assessment, including intelligent tutoring systems that allow text to speech, translation, personalized learning, and inquiry-based learning.”

Teachers may also find this example rubric helpful (or at least interesting).

Tip: Be wary of Canva GPT quotations

I saw Canva’s custom GPT promoted as a tool for quickly making social media posts, and since I already like Canva (enough to pay for a Pro account), I decided to test it. The tool seems like automation for templates, but unfortunately, I was not impressed.*

I asked it to create an educational Instagram post about semicolons, since that should be a relatively straightforward request.

As you can see, the design itself had problems, such as putting the name of the person quoted twice. But that’s easy to fix: you can click on the image from inside Canva GPT and go to Canva, where you can quickly edit the image.

A much bigger problem is that Amy Bleuel doesn’t seem to have ever said the quotation. I tried multiple Google searches to try to find it, and I’m 95% sure it doesn’t exist. She said similar things, but not this.

What worries me is that these AI-generated quotations soon will appear on the web, and then in Google searches, and then it will be harder to tell which quotations are real and which aren’t.

This is actually already a problem with quotations online and is why I love the Quote Investigator. But just like I worry about the problems with search results, which I talked about in the last newsletter, I fear fake AI-generated quotations are going to flood the internet.

News

Artists can now actively poison AI

A team at the University of Chicago just released a tool called Nightshade that makes invisible changes to digital images. The selling point is that these changes “poison” AI models that try to use the images as training data. You may feel like you’ve already heard about this, but Nightshade is a new tool released by the same team that made Glaze last year. The difference is that Glaze is a defensive tool that prevents AI models from using the image to train models in a way that allows them to mimic an artist’s style, and Nightshade is an offensive tool that actively harms AI models.

This part of the description made me laugh: “While human eyes see a shaded image that is largely unchanged from the original, the AI model sees a dramatically different composition in the image. For example, human eyes might see a shaded image of a cow in a green field largely unchanged, but an AI model might see a large leather purse lying in the grass. Trained on a sufficient number of shaded images that include a cow, a model will become increasingly convinced cows have nice brown leathery handles and smooth side pockets with a zipper, and perhaps a lovely brand logo.” — The Nightshade Team at the University of Chicago

As you might expect, art people seem to love this idea. On Threads on Friday, an art enthusiast said he was going to treat and upload a “few hundred” images to free stock photo sites over the weekend so they get used all over the internet.

Those who are less enthusiastic claim that although it’s fine to modify images on the web, it would be “highly illegal” to actively upload images you know would harm AI models. (Others have commented on the irony of calling it illegal when Nightshade is meant to deter what many artists consider theft.)** Further, there is some discussion about whether the tool actually works.

Regardless, I suspect this is the beginning of a technical back-and-forth, with AI companies finding ways to counteract tools like Nightshade, and artists and their supporters finding new ways to thwart their systems.

Multiple people have asked if something like this would be possible with text, and I have no idea! But I do wonder if someone could develop a font that somehow thwarts AI training models by making the letters uninterpretable or by adding invisible characters (☠️ Helvetica Doom ☠️). It’s probably more complicated than that.

It’s also important to remember that tools like Nightshade only apply to newly created content. The vast databases of existing images, text, and audio are, of course, still available.

Nightshade overview.

Free Summit: AI for Writers

Amy Frushour Kelly posted about an interesting four-hour online event on March 6 called the “AI for Writers Summit.” I have signed up for it.

From the website: “Everyone, from large enterprises to freelance copywriters, has access to affordable AI technologies. It’s the understanding and application of AI that will differentiate you and your company moving forward.

You can already 10x your content output—ads, articles, blog posts, captions, emails, newsletters, social media posts, transcriptions, video scripts, websites, and more—with today’s AI, and it’s only going to get more powerful from here!

Vision, innovation, and speed matter more than ever. 

If you are a writer, editor, or leader of a content team, it is essential that you take action to understand and apply AI.”

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.

^{* This is one reason there is no “Grammar Girl” GPT. I’ve played with making custom GPTs, and I don’t think they are remotely consistent enough, and I wouldn’t want to attach my brand to one. Still love ya, Canva, but not the custom GPT.}

^{** I’m not a lawyer and don’t know whether it is actually legal. This is not legal advice.}

^{Written by a human.}