I Don’t Want to Interact With Stochastic Parrots

My AI policy is "I don’t want any kind of interaction with that thing." And this should be all you need to know.

I don’t care if you call it AI, LLM, chatbot, or whateverGPT. I simply don’t want to interact with it in any form. If you interact with me, make sure to never send me any output from those nor to use it as a "source" to justify one of your arguments.

I’m one of those people deeply interested in all the extraordinary stuff other humans have to say. Unfortunately, life is already too short to hear them all. I need to make a lot of choices. That’s why I will not tolerate any attempt to steal the little time I’ve on this planet by making me interact with stochastic algorithms trained to produce tons of "average," non-human content.

The direct consequence of my choice is that I will refuse to interact with humans who take pride in making themselves a meat interface between me and the algorithm they’ve chosen to replace their free will.

I will not read your website, I will ignore your emails and I will happily walk away from real-life conversations as soon as it is clear that I’m not talking to a fully sentient Homo sapiens sapiens but to an empty shell who has happily delegated their cognitive abilities to an external proprietary tool designed to be some sort of "random average garbage."

I will not read the text

The very essence of writing is what makes you think. When you write, even something as mundane as a short email, you transform a bunch of contradictory neural stimuli into an ordered thought.

If you don’t care to write it, there’s no reason I should waste one single second reading it. The effort should be on your side first.

I don’t mind mistakes, typos, or badly formed sentences. In fact, I subtly appreciate them when we interact because they are part of you, they tell a lot about yourself. I’m not a literary agent, I will not judge your way of writing: I’m there to exchange ideas, knowledge, and experiences.

If you find yourself typing a prompt to express your idea, then just send me that prompt. I’m not interested in what an algorithm would do with your prompt, I want to know what you asked in the first place. I care about you, the human.

You may find it hard to express yourself, to write what you think: remember that the less you practice, the worse it will be. Prompting might feel easier. It does not make it right.

If it is from an algorithm, it is not you. Only you can write what’s inside yourself.

I will not look at the pictures

Generated images. Seriously. They all look hideous.

When I spot one in a given context, I immediately have a feeling of cheapness, bad taste, and poor culture. Like this fancy restaurant of a luxury hotel I saw in a big European city. It had a huge AI generated picture on its main building: a kid eating malformed fries with his six-fingered hand and putting them in his skull through his cheek. Or this other restaurant which had a very nice menu, full of pictures of Italian cities. Until the kid’s menu which, was shockingly out of place and totally unnecessary. It also showed a kid awkwardly eating a pizza even though there was no pizza on the kid’s menu.

If you don’t have the resources to pay for an artist, you probably don’t need a picture at all. If you are making a poster, write what is important to know instead of burying it in a gloomy generated picture. If you write a blog post, I will not read if it is illustrated by generated pictures.

A website illustration, if any, conveys an emotion, a feeling. I sometimes spend more time finding a good illustration for a post than writing it. If I don’t find a satisfying picture, I don’t put any.

The first feeling conveyed to me by a generated picture is an uncanny unease, an obnoxious emotion that immediately makes me want to leave. It also tells me enough about the author not to want to know more. It might be a kind of overreaction. But as I said, there’s too much to read anyway so it’s a good first filter.

No, that picture is not pretty. No, it is not funny. It doesn’t share any human feeling, any artistic expression. You like it because you had fun generating it. But it is wasting my time.

I will not run the code

I don’t want to run generated code on my computer.

Good code is knowledge. Even in a purely productivity-oriented business, every line of code should reflect the choice of an employee knowing the business and choosing a way to solve a business problem.

Reality is, of course, different. I’ve worked with enough different codebases to know that proprietary software is simply badly coded. Each line is written assuming "I will not be there anymore to deal with it in a few years." But at least we know that each line, as bad as it can be, was added by a human for a reason. LLMs move the needle from "no need to care in a few years" to "no need to care right now." You cannot assume anything about generated code. Maybe it was simply hallucinated and, by chance, it compiled.

Which means that, as soon as generated code is added to a project, the whole project becomes unmaintainable by humans. You can’t understand the code because there’s nothing to understand in the first place!

It follows that the whole project now needs LLMs for any change which, will make it even worse from a human perspective. We end up adding more and more lines of code nobody understand where the problem has always been "There are too many lines of code and too few people understand them."

I’m convinced that accepting generated code will lead most projects to become broken beyond all hope of repair in only a few years. "Carefully reviewing" will soon not be enough. Unfathomable code will eventually slip in and gradually destroy the whole project. I predict that the projects too important to be left to rot and die will take decades to recover from those couple years of "assisted productivity."

We need to remember that what we call "code" is merely a specification of the software we want. Those specifications will be given to the compiler to make the real software. The role of a coder is to take vague, imprecise, and contradictory specifications from a manager or customer and then make sure to remove any ambiguity. This could be done by clarifying with the customer or by silently making choices. History and mathematics have demonstrated that humans cannot make unambiguous and non-contradictory specifications without using a formal language designed expressly for that purpose. Natural languages simply cannot remove ambiguities and this may be an evolutionary feature.

Removing the coder and replacing them with a tool designed to pretend to agree to anything said could never produce good, maintainable software. It may produce a small throwaway script. It may find security holes more efficiently than fuzzing. But it will never be able to build something useful and reliable for the long term.

Which is what I expect from software.

I want to be able to trust software enough to invest time in learning it. That has always been one of my main arguments for using open source: if a free software is good enough, I trust that there will always be people to maintain it or to fork it if needed. If I have a problem with a free software, I can always look at the code to understand what the humans who designed it wanted to do. I may also modify it. This is not theoretical: I’ve done that with countless pieces of software in the last 20 years. I rely on it!

For that reason, I will carefully avoid running any openly "vibecoded" tool, and I’m trying to focus my efforts on projects that actively try to distance themselves from the current bubble.

I will not listen to the discussion

"I’ve asked ChatGPT and…"

That’s it. No need to say more. All those years speaking about "fake news" and "checking your sources" just to blindly repeat a random series of letters assembled by a stochastic algorithm? Seriously?

I don’t care about your credentials, I don’t care about your experience. With a couple of words, you gladly announced that you are obliterating them and refusing all forms of free will. Goodbye!

That’s not very friendly, I know. I never pretended to be. And I didn’t start it.

I will not grade the student report

Across the last years, I’ve not managed to find one single student who astonished me with their use of a chatbot. In every single case I’ve met, the use of a chatbot made my student perform worse. Sometimes spectacularly like, this student who contributed to a project on Github but, when asked, could not even tell me what the project was about.

On the opposite, forcing the student not to use any AI makes them better every single time. No exception. It also vastly improved their self-confidence.

• Giving University Exams in the Age of Chatbots (ploum.net)

It is not a matter of "learning to use the tool". You cannot learn to use a proprietary stochastic tool which can be changed at any time. You don’t need to "learn to chat" (and if you do, not in my course). What the "tool" is really doing is making you lose the essential ability to learn, making you lose your self-confidence. As John Scalzi points out: playing Guitar Hero is fun. But even thousands of hours of playing Guitar Hero will not enable you to play "Happy Birthday" on a real guitar.

• Generative “AI”: The Guitar Hero of Creativity (whatever.scalzi.com)

I consider that my students are in my class to learn. Chatbots are actively preventing any kind of learning. That’s why, from now on, I make it a strict policy that anything my students are submitting should be entirely produced by them.

It can be short. It will be imperfect. It will be full of mistakes. It may even be wrong. The report will have fewer bullet points and emojis. But it will be yours, and that’s what I want to see.

I will not technologically "enforce" this rule. Cheating has always been part of university. If you are smart enough not to be caught, you probably deserve to pass. But it will probably require a lot more effort on your side than simply writing your own goddamn report with your bloody fingers on a keyboard.

I will not judge what you do in private

AI/chatbots/LLMs are known to have a terrible impact on your brain. They are bad for you, for the whole economy, for the political atmosphere, and for the environment. They consume huge resources, destroy the whole Internet and our ecosystem.

But they are addictive, I get it. They make you feel productive (which, studies have demonstrated, is largely a feeling). I consider them a kind of drug.

• AI Isn’t Breaking Work. It’s Already Broken. (calnewport.com)

They may also be useful when you are forced to accomplish a stupid one-time task you don’t want to do but is required. We all have to deal with our share of bullshit jobs, I get it.

Using AI is like a shameful drug addiction or a weird sex fetish. If you want to use it privately or between consenting adults, it’s up to you. As long as you don’t brag about it, I will not judge.

Just make sure that I don’t have to interact with it.

• Human made logo by Matt Kindt (commons.wikimedia.org)

Edtech psychosis is everywhere lately. You have Randi Weingarten, president of the American Federation of Teachers, partnering with AI companies one month and then denouncing AI a year later. You have New York City Public Schools banning then unbanning and rebanning AI in their schools over a span of three years. You have the Los Angeles Unified School District promoting their custom AI chatbot one year and then banning AI two years later. You have edtech insiders all over LinkedIn excited about technologies way out of proportion to any evidence that they support student learning.

Edtech psychosis flourishes wherever a giant tech company’s free cash flow meets education’s threadbare operating budgets. Edtech psychosis flourishes in people who are more impressed by a novel business model or a speculative technology than by good teaching or a well-designed curriculum. Edtech psychosis flourishes in people whose pedagogical convictions are too weak to withstand a wink from whatever new technology walks by.

If your job is to support students, teachers, or schools, then edtech psychosis is fatal for your project and you should hope for an inoculation. Edtech psychosis will stunt your understanding of the reality of schooling. You’ll imagine students as little empty hard drives or young disobedient employees. You’ll imagine teachers as frantic CPUs, desperately trying to write different bits of information onto every student. You’ll imagine most kids learn as you did—well-supported and self-motivated. You’ll imagine all kids are as eager as you are to work and learn by themselves on a laptop, as eager as you are to learn what a large language model might say next.

How do you inoculate yourself against edtech psychosis?

There are people who succumb to edtech psychosis and people who seem immune to its debilitating power. Immunocompetent people understand the “grammar of schooling”—the goals, constraints, routines, and aspirations that power the people who work in schools. The immunocompetent are teaching classes, observing classes, watching videos of classes, or, at minimum, reading testimonials from teachers all as often as they can. The immunocompromised do none of those things. Instead, they congregate on LinkedIn like kids at a chickenpox party, each one wondering in their own perplexed way why people aren’t as impressed by artificial intelligence as they are.

How I’m inoculating myself against edtech psychosis this year.

As of last week, I have successfully begged my way into a fifth grade public school classroom here in Oakland, CA, where I’ll work weekly this school year. I was trained as a high school math teacher and have since taught middle school but, until now, I have only experienced elementary school as a parent. This will be an adventure. Next week I’ll share some early observations.

What is your inoculation plan?

Time in classrooms keeps my imagination pointed at the actual rather than imagined problems of kids and teachers. It keeps me inoculated against edtech psychosis. Let me know your own inoculation plan in the comments or by direct message. Email a local school. Meet a teacher at a weekend farmer’s market. Ask if you can help. Get yourself fingerprinted. Get yourself a name badge. Become a substitute teacher and get paid to get inoculated. One hour per week will completely change your outlook. Let me know after you have received your inoculation, please, so I can stop worrying about you.

Upcoming Presentations

I am keeping busy this fall and hope to see you at any of these conferences.

• September 14. Online. Keeping High School Students Engaged in Math: Calculated for Success

• September 16. Kansas City, MO. The Power of Problem-Based Learning.

• September 17. Boston, MA. A panel on artificial intelligence & emerging school models.

• October 9. Portland, OR. Northwest Math Conference.

• October 16. Online. A keynote for the Australian Association of Math Teachers.

• October 30. Denver, CO. The National Council of Teachers of Mathematics annual conference.

• November 5. Palm Springs, CA. California Math Council’s annual conference.

Odds & Ends

¶ Los Angeles USD and New York City PS both enacted sweeping bans on artificial intelligence in the last couple of weeks. Over on LinkedIn, edtech operators are either trying to put on a brave face or they are fully rending their garments. My commentary here is brief: with artificial intelligence, you spent four years promising baby while delivering bathwater and too many people noticed. Next time: consider underpromising, overdelivering, or both.

¶ Linda Jacobson at The 74 describes the ways AI in education has underdelivered in her round-up of recent research. Wise words in the article from high school history teacher Wendy Graham: “Kids don’t do their best for robots; kids do their best for people.”

¶ I appreciated this Justin Reich column which notes that the absolute best case hypothetical educational website lives inside a browser which lives inside an operating system which lives inside a device. The best case hypothetical educational website controls none of those outer layers and each one of them has goals for users that often run counter to the goals of that absolute best case hypothetical educational website.

¶ New York Times reporter Natasha Singer has a quote from Brad Smith, Microsoft’s president, that we should inspect closely.

Brad Smith, Microsoft’s vice chair and president, told me the dynamic began because public schools often seemed ill-equipped to teach crucial modern subjects and prepare students for the ever-evolving demands of the job market. He also suggested I consider where schools might be without support from corporations like Microsoft.

“Philanthropy in the private sector is doing this work because the public sector has failed to do what it’s supposed to do,” Mr. Smith said. He allowed that the situation wasn’t ideal. “It’s filled the gap out of necessity,” he said. “It’s not the way we should want public education to work.”

Of course public education doesn’t operate the way you want it to, Brad. Private industry has always wanted the public to subsidize its employee development costs. The public simply doesn’t share either your opinion about the point of school or your fiduciary duty to Microsoft shareholders.

¶ My local district here in Oakland, CA, is contemplating screen time restrictions. School board member Rachel Latta came on board when she walked into a classroom and noticed the effect of technology on the class.

[Latta] said she has sometimes visited schools and seen entire classes of students not speaking to each other, but working independently on laptops with headphones on, which struck her as odd.

Yes, it’s odd! It’s very odd. We build and operate these expensive buildings called “schools.” We fill them with caring adults and we beg kids to come and then when kids arrive we say, “Whoa whoa whoa, let’s not have too much together time now! Here are your laptops and headphones.” People seem to have substantially less patience nowadays for technology that tears down rather than builds up a community.

Students who use AI to help them study tend to perform worse at school than those who don't, according to data from a global OECD educational report. The situation is more complex than it sounds though, with certain types of AI use giving learners a slight boost, especially among students taught to critically assess how well the AI tools perform.

The OECD's Programme for International Student Assessment (PISA) takes data from countries around the world every few years. This year's study, based on data collected in 2025, is the first to be carried out since AI use went truly mainstream. It tests 15-year-old students in science, math, and rea …

Read the full story at The Verge.

A generation raised on the technology is discovering there are shortcuts it doesn’t want to take

I’m not a fan of having students refer back to notes they have written. Wasn’t a fan when I taught high school, or when I taught middle school, or now in elementary.

First, I have a disclaimer. Writing things down is great for learning. Putting ideas into one’s own words, summarizing ideas, trying an example, putting pencil to paper. I am very much in favor of this! I have students keep a composition book on their desks. I often have students open it and do a bit of writing to summarize something I want them to learn. We do it in every subject. Writing is good.

What I don’t do is ask students to use that thing they have written down as a reference, as something they are expected to find and help them do things in the future.

If there is a resource I want students to have — a formula, a definition, a list of things to include in a paragraph, a map, a worked example, anything — I provide it to them when they need it.

The Pitfalls

Here is a list of things that happen when teachers ask students to take and refer back to notes which render those notes less useful:

• A student takes notes incorrectly.

• A student takes notes illegibly.

• A student takes notes but takes so long that they miss critical instructions for what is happening next.

• A student takes notes by copying from the board without thinking about what they are writing.

• A student is absent and does not have the notes.

• A student is present but chooses not to take notes.

• A student is absent and gets the notes from some other means, but the teacher spends so much time making sure absent students have all the notes that they take away instructional time from other priorities.

• A student takes notes successfully but cannot locate them when they are needed in the future.

• A student takes notes successfully and locates the notes when they are needed, but they have spent so much time flipping around looking for what they need that they lose valuable learning time.

• A student takes notes successfully but does not know which notes are relevant to a future task.

• A student takes notes successfully, locates the correct note, but does not know how to apply that note to the task at hand.

It’s a minefield.

In my class we write things down all the time. Writing things down is good! Students should write things down. But I don’t stress about whether the things that are written down are useful for students to reference in the future. A lot of what I do might look like notetaking! Try this problem. Write a sentence with this sentence starter. Write down an example of a simile from the text. That’s good stuff. But what I don’t say is, “look back at your notes to help you figure out how to do this thing.”

Opportunity Cost

The argument I’m making here is fundamentally about opportunity cost. I’m not telling teachers that their notetaking routine is broken. It probably works for most students! I’m arguing that the time you put into a notetaking routine might be better spent on something else.

All of those concerns I listed above are possible to solve. Possible, but time-consuming. And time is precious. I could make sure every student has usable notes for every task we do. I could ensure that my future tasks are perfectly aligned so their notes are always useful. We could organize notes. I could have students practice looking back for different pieces of information.

But that is SO. MUCH. TIME. Time organizing notebooks, making sure every student has the notes, getting notes for absent students, flipping back through notes looking for a piece of information. Time that could be used for...well, anything. So many other things that help students learn. More practice, more feedback, more reading, more writing, more questions, more this, more that.

There are lots of pieces of information that could be useful to students in the future. Maybe it’s a formula for a math problem, or a structure for a piece of writing, or a definition, or a map, or an example of something for them to reference if they get stuck. I give students that information. The format varies: often it’s on the board, nice and big, easy to find. Or on an anchor chart that I can leave in the back of the room when we don’t need it, and bring to the front when we do. Or on a piece of paper I give to students. I make sure it’s clear, relevant, and helpful.

Some of you are saying, “but I found my notes so useful for blank and blank and blank.” Cool! I have students write down lots of stuff. They could choose to use those notes if they like. I don’t ban students from taking notes. It’s just not something I require.

Is Notetaking a Skill?

A classic result in cognitive science is that when students refer back to their notes, they overestimate how much they have learned. Looking back at notes is consistently inferior to other, more active ways of studying.

There’s an illusion of learning at play here. A student might look back for something, find it, and think that must mean they have learned it. It might look to a teacher watching students diligently take notes like those neat notes mean learning has taken place.

One risk with notes is that the goal of education becomes to have lots of well-organized notes and resources to reference. To me, the goal of education is for students to have lots of well-organized knowledge in long-term memory. That doesn’t mean references are useless, but it’s a different goal. My experience is that it’s much easier to move information into one’s notes than into one’s long-term memory, and it’s tempting to substitute one goal for the other.

Another common complaint here is that notetaking is a skill, and we should be teaching students to take good notes. I’m not convinced of that. I do want to teach students good study habits. I often tell students that in class we write about what we are learning so that learning is stronger. Putting ideas in your own words, or quizzing yourself, or trying another example on your own are great ways to learn. I’m not convinced that “take thorough notes and reference them in the future” is especially good learning advice.

I’m also just not convinced that notetaking is a generic, transferable skill that we can reliably teach students. I think many students will figure out notetaking skills on their own, whether they are taught or not. Many more won’t develop notetaking skills despite a lot of instruction. Plenty of students have a dozen different teachers who try to teach them how to take notes, and don’t actually ever develop that skill.

A common argument here is that in college, students will have to take notes and learn much more independently. To be blunt, most notetaking routines are about compliance. They’re about copying things down, then finding that thing you wrote down later to prove that you’re a diligent student. That’s not what college is like. Students who show up and try to write down everything the professor says aren’t going to learn much. A student who has practice summarizing ideas in their own words and being an active participant in their learning is going to learn a lot more, whether or not they ever refer back to their notes. Those are the habits I try to model for students.