Marc Watkins

A time when you should maybe not use AI:

My daughter starts back to school tomorrow in the 6th grade and the school used an AI voice over video with slides likely generated as well to talk about how her school was all about relationships.

Welcome back?

I'm pretty sure the reason they're releasing their first open model since GPT-2 is because of Deepseek and Kimi K2. China is set to dominate the market with open models. Interestingly, Zuck appears to be backing off open sourcing future versions of Llama. I guess time will tell.

In the process with SM and will likely start with AI once people realize the issues teens are using it for go outside of school work.

Sure, but it is even more basic than that. I don't sense many people understand that ChatGPT is just the UI and that the models continue to change under the hood. The average user associates GenAI with ChatGPT-they don't understand alternatives exist or the models under the UI change rapidly.

Indeed. My point: do you realistically foresee Microsoft, Apple, or Google allowing a 3rd party LLM to be easily loaded onto a device using their OS? By default the first on device models have been proprietary to those providers and constrained. What incentive would they have to allow it?

Linux is also 100% free but adoption is low among OS users because it takes time to learn, time to set up, time to troubleshoot, time . . . AI is marketed on efficiency and ease of use. Many people still don't know what a web browser is. They simply click "on the thing bundled with the other thing."

There isn't a viable market to corner for on device LLMs right now because it doesn't exist. We'd need to see vast improvements in local models and social adoption among non-nerds for a real market to form. There are massive factors involved here, like human behavior and preference for easy use.

Age Verification Is Coming for the Whole Internet Around the world, more users are being asked the same question: Can we see some ID?

I don’t disagree but it is happening and is a shift not viewed as surveillance (which it is) but as enhanced safety for minors. Really difficult to argue against it with child safety as framing nymag.com/intelligence...

I doubt this will be viewed as anti AI policy regulation. ID/verification will be sold as keeping minors safe from harmful synthetic content. The damage to companies comes from the 2nd order effect that few adults are willing to share their ID to access a service. That’s a behavioral issue.

Generative AI is a massive public experiment, so too was social media. We’ve seen the bad parts with addiction and isolation with SM and we’re now starting to see it with AI among certain vulnerable groups. A public reckoning is coming. It’s just a matter of when and how far it goes.

If teens are increasingly chatting with AI as virtual friends and seeking relationships with it then many people we see this as a threat to mental health and blanket requirements like ID verification are a real possibility. It will kill AI adoption and VC funding b/c new users will vanish.

On device won’t be immune from guardrails. Apple along with any provider would have to follow age verification if passed into law to activate an on device LLM, and again, I don’t see people giving their ID to access a service.

Use search, yes. Be optimized and updated routinely to ground results to avoid errors, no. People want AI as a service, they don’t want the challenge of downloading and running a model. If the service aspect was nerfed, I’d fully expect hundreds of millions of people to stop using it.

On device LLMs aren’t competitive with reasoning models or agents. In my own testing a local reasoning model hallucinates wildly. They need grounding with search to be effective. And the shelf life of a local model is measured in months right now in terms of market pressures. ChatGPT rules the day.

Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions Our new report finds that teens' AI companion use is commonplace, complicated, and sometimes concerning.

Forget sexy mode, Musk doesn’t have guardrails for Grok. You can chat all day with conspiracy mode and dozens of other 3p sites offer similar. The surveys out of US and UK indicate teens are using AI for companionship and therapy in high rates. www.commonsensemedia.org/research/tal...
+

Marc Watkins Marc Watkins is assistant director of academic innovation at the University of Mississippi, where he directs the AI institute for teachers. He writes regularly about generative AI in education on his ...

Generative AI is now integrated into most of the apps we use each day. It’s very hard to tell when a search result it generated or a summary is in some apps. I wrote a column on the Chronicle about AI www.chronicle.com/author/marc-...

We may see policy indirectly impact tech adoption through ID verification. Most states now have it for porn sites and there’s discussion to bring it to social media. I could see AI next. Very few people want to share their ID to access a service. That could cripple AI adoption if it is applied.

The new “Study Mode” for ChatGPT turns it into a Socratic tutor and is just a prompt. The implication is OpenAI could have released this years ago—a version of ChatGPT that wouldn’t complete your homework for you.

Let that sink in. OAI could have nerfed ChatGPT with a few well organized phrases.

Some Things Need to Be Grown, Not Graded, and Definitely Not Automated The very real frustration teachers have over rampant AI usage in classrooms is growing.

open.substack.com/pub/marcwatk...

As the fall semester approaches, we’re bound to hear more talk about AI policies, what new skills students will need, and how to curb AI misuse. We’re also likely to hear renewed calls for faculty to redesign assessments. A year ago I wrote a series about how AI impacted learning and revisited it.

Advice | How to Grapple With the AI Already on Your Campus Three steps any faculty member can take to understand which AI features are now embedded in applications you use every day.

My latest for the Chronicle -- how to navigate the AI already present on your campus. Now that ChatGPT will be directly integrated into Canvas, I think it is time we have a much deeper conversation about how quickly AI has been integrated into existing applications www.chronicle.com/article/how-...

OpenAI moves deeper into schools via Canvas deal Teachers will be able to build custom chatbots to manage classrooms.

Posted without coment www.axios.com/2025/07/23/o...

AI Awareness Starts with Time Time is one of the luxuries every teacher searches for within their classrooms.

Time dictates so much of our engagement with students in higher education. Here are some practical ways you can talk to them about AI that don't feel overwhelming to you.

open.substack.com/pub/marcwatk...

My feeling is textual feedback is easy to dismiss, but having a model talk with you through feedback, persuade you, even flirt with you is going to become addictive. We think that faculty freakout is strong now, image when folks realize students have a version of HER.

Anthropic is quietly pushing ahead with the economic disruption discourse and I don't think they believe it is hype. They're now directly funding academic research, though the CFP reads more broader than just economists. docs.google.com/forms/d/e/1F...

A.I. Videos Have Never Been Better. Can You Tell What’s Real? (Gift Article) In this quiz, you can guess whether these videos were made by A.I.

This isn't a challenge. The videos may appear realistic, but the audio filter Google uses for dialogue in Veo 3 is stunted compared to human speech. It's like listening to someone talk on an airplane. It is an easy tell . . . for now. www.nytimes.com/interactive/...

I have many snarky things to say about this philanthrocapitalist propaganda sponsored by the family that brought you food deserts.

But let’s keep it simple.

Corporations have always used tech bubbles to disrupt labor solidarity & smuggle labor intensification under the veil of efficiency.

Who Owns the AI Dividend? There is a growing divide among those who use AI and those who do not.

The words “returns,” “savings,” “gains,” and “impact” appear repeatedly throughout the Walton Foundation's Unlocking Six Weeks a Year With AI report. It sounds like a technology ad promising an amazing return on investment, not an educational report about pedagogy.
open.substack.com/pub/marcwatk...

The AI Dividend: New Survey Shows AI Is Helping Teachers Reclaim Valuable Time Gallup-Walton Family Foundation poll finds teachers who regularly use AI tools save the equivalent of six weeks per school year

The Walton foundation report about AI usage among K12 teachers confirms most teachers who use AI do so to save time, saving some teachers 6 weeks a year.

Your kid may not be allowed to use AI in school, but his teacher is becoming a power user.

www.waltonfamilyfoundation.org/the-ai-divid...

Thanks, Jane!