The Pedagogy of Presence: Restricting ChatGPT-6 Astra to Protect and Cultivate Student Thinking
When AI can research, reason, cite sources, and produce nuanced essays, teaching writing becomes less about the final product and more about protecting the thinking required to produce it.Photo by Jonathan Kemper on UnsplashA little over a year ago, ChatGPT-5 stopped me cold. ChatGPT-6 Astra leaves…
When AI can research, reason, cite sources, and produce nuanced essays, teaching writing becomes less about the final product and more about protecting the thinking required to produce it.Photo by Jonathan Kemper on UnsplashA little over a year ago, ChatGPT-5 stopped me cold. ChatGPT-6 Astra leaves this high school history and journalism teacher wishing that were still my biggest concern.Since ChatGPT’s first iteration in 2022, I had believed that I could manage generative AI in my classroom. After all, I knew my students and their writing.I was familiar with how they constructed arguments, where they struggled, and the unique stamp each put on their prose. Then I witnessed ChatGPT-5 mimic those voices with frightening accuracy.That experience changed my classroom. I delayed assigning substantial take-home writing until I could familiarize myself with each student’s unique style. I moved more writing into class, where I could watch them think. I wanted to see the process, not simply collect the final product.I also added culminating one-on-one conferences, not to “catch” academic dishonesty, though that did happen on occasion, as when I questioned why students cited a paper about World War I while writing about something entirely different in World War II (an AI hallucination).My goal, which I wrote about in a recent Edutopia post, was to better understand students’ learning processes and offer live, immediate feedback rather than written feedback, which could be misconstrued as “mean” or “aggressive” (a different issue entirely).Then, on Thursday, OpenAI announced “A New Generation of Intelligence” with its latest release, Astra, which is slowly rolling out across various subscription tiers.https://medium.com/media/1e1473496ae69188a03f5a113f550a79/href“It can conduct online research and draft summaries in your email or in your document editor,” the company’s release announcement says. “It can analyze scientific data, generate plots, create a website, and…make sure all the features on that site work.”And that’s to say nothing of the cognitive offloading it offers students: before they’ve even learned to walk, Astra will prop up their unformed minds and do the heavy lifting for them at astounding speeds, much like WALL-E’s hoverchair-bound passengers, only for them to crash and burn the moment it’s unplugged, or the moment they’re asked to think for themselves.https://medium.com/media/ddf9894c8759919fa44e5d2fc17c8d34/hrefVia my Plus subscription, I happen to have access to GPT-6 Astra under the “Work” tab, but only up to the “Medium” reasoning level, not “High” or “Highest.” Even that was enough to make my blood boil.And after just one task, I had to pay an additional $30, plus tax, to complete my experimentation for this article without waiting several hours between engagements. Technocapitalism at its best, but in this case, I fed the OpenAI leviathan for the good of educators everywhere.I asked Astra to provide exemplars for a take-home essay topic I assigned at the end of last academic year, and I was blown away by the results, which contain (as far as I can tell) accurate footnote citations and clear, analytical writing.I also instructed Astra not only to cite directly from the resources I provided to students, but also to conduct its own independent research. Of note, I offered Astra no additional instructions or refined commands beyond what I provided to students in the assignment and the rubric.The biggest stride forward from ChatGPT-5 that I notice, however, is the AI’s ability to embrace nuance. I was stunned, for example, by the first part of the conclusion to its first exemplar response: “Truman’s decision should consequently be criticized without pretending that he possessed a painless alternative or perfect foresight.Japan’s leaders bore responsibility for aggression and continued resistance; American troops and occupied populations had urgent claims to protection. Those truths justified determined efforts to end the war.They did not justify treating diplomatic uncertainty as evidence that negotiation was useless, military value as permission to devastate a population, or the first authorization as sufficient warrant for a second attack.”Agree or disagree with the conclusion, this is human-like, nuanced reasoning that makes this history teacher wonder what the technology could do at its highest capability level. My biggest (and growing) concern, though, is how much responsibility students can now delegate.Generating an answer already lets them bypass the writing itself. An agentic system can take over the rest: finding evidence, organizing the project, producing files, and checking whether the result meets requirements.If you are interested, I left several annotations in the first AI-produced essay, as if I were writing them to a student, and I welcome your own thoughts.It’s early in Astra’s deployment, but anecdotally, I was super impressed by the lack of hallucinations, though recent reporting, including from Fortune, suggests OpenAI’s officially published hallucination figures have shifted more than once since launch, so for now I’m leaning on my own reading of the essays.https://medium.com/media/9a276061b5a49aa89ab293ba9c52b6ce/hrefBut clearly, the biggest question writing teachers have to face now is why it’s even worthwhile to teach writing anymore, when AI can write so well, and with such nuance. I can only answer that writing is thinking, and just because a machine can do something doesn’t mean that it should.Otherwise, we are doomed to foster a generation of apathetic minds, who care too much about what machines can do for them and too little about what they could and should do for themselves. I’ll leave it to you to decide where to draw the line, but as far as my classroom teaching goes, Astra isn’t allowed.To the critics who say I am ignoring the inevitable, and that I should instead teach my students how to use LLMs and now agentic AI, I can only say that we will certainly discuss the technology, its capabilities, and its implications for, well, just about everything.In fact, I am even assigning my AP United States Government and Politics students the first chapter of Jill Lepore’s fantastic new book, The Rise and Fall of the Artificial State. But just as students cannot drink until they are 21 or run for president of the United States until they are 35, there is a reason, I believe, to ask students to wait until they have developed sufficient maturity and skills before subjecting themselves to Astra and its future iterations.And for those who say I am hurting my students’ job prospects in a world where AI will continue to be essential, I can only say that I see my job differently.My role is not necessarily to equip students with whatever skills employers may covet next — which is an impossible feat anyway — but to help foster inquisitive, independent minds capable of thinking, questioning, and reasoning for themselves, so that however nuanced and intelligent AI becomes, my students will be equipped to question it and push back where appropriate.That, with every fiber of my being, I believe to be an everlasting asset.This story is published under the Generative AI publication. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!The Pedagogy of Presence: Restricting ChatGPT-6 Astra to Protect and Cultivate Student Thinking was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI