Jane Rosenzweig
LightNews LightNews
janerosenzweig.bsky.social
Jane Rosenzweig
@janerosenzweig.bsky.social
3.7K followers 1.2K following 370 posts
Writing in the age of AI stuff
Newsletter: writinghacks.substack.com
Also curate theimportantwork.substack.com
Posts Replies Media Videos
Pinned
janerosenzweig.bsky.social
Jane Rosenzweig @janerosenzweig.bsky.social · 19d
I have a short piece about what I've learned from teaching my AI course in this @chronicle.com forum on how AI is changing higher ed, along with some very interesting reflections from others, including a lovely piece on why writing matters by @zey.bsky.social . www.chronicle.com/article/how-...
Opinion Forum | How AI Is Changing Higher Education
The technology is reshaping every aspect of university life. Fifteen scholars weigh in on what happens next.
www.chronicle.com
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
I took the K-12 student certification to see how it works. Sample question.
question from multiple choice exam: Gemini can make literature, like a play by Shakespeare, more accessible by explaining it in simpler language. True or false. The correct answer according to Google is "true"
November 24, 2025 at 12:03 PM
janerosenzweig.bsky.social
I took the K-12 student certification to see how it works. Sample question.
question from multiple choice exam: Gemini can make literature, like a play by Shakespeare, more accessible by explaining it in simpler language. True or false. The correct answer according to Google is "true"
November 24, 2025 at 12:03 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Question 5 of 8 29893472 Fill in the blank: In order to generate helpful outputs, _____ are trained on vast amounts of text, which teaches them to identify patterns among words, concepts, and phrases. large language models audience members text-to-speech solutions prompt iterations
November 23, 2025 at 2:15 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
This is fascinating because it assumes not that teachers won't know how to use AI but that they do not understand basic concepts.
Image of text that says "Make it stick. Here's a quick activity to review a key concept from this lesson." Then it shows a prompt that says "I am a middle school history educator" and has a multiple choice quiz to answer whether that prompt is defining the goal, formatting the result, choosing a role or persona, setting restrictions, specifying audience."
November 23, 2025 at 2:12 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Checking out some of Google's AI training for teachers and this seems like a very easy quiz or else a strange understanding of teachers.
A quiz that asks teachers to figure out if "generate email templates" is "time-savings" "differentiation" or "lesson enhancement"
November 23, 2025 at 12:54 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Pretty amazing AI-generated item available on Amazon.
boy in front of a christmas tree holding a calendar that says "Hannukah Countdwand, 24 Days of Miracle 2025" and text that says "Hanukkah Countdown Calendar 2025"
November 23, 2025 at 12:23 AM
janerosenzweig.bsky.social
Checking out some of Google's AI training for teachers and this seems like a very easy quiz or else a strange understanding of teachers.
A quiz that asks teachers to figure out if "generate email templates" is "time-savings" "differentiation" or "lesson enhancement"
November 23, 2025 at 12:54 AM
janerosenzweig.bsky.social
Pretty amazing AI-generated item available on Amazon.
boy in front of a christmas tree holding a calendar that says "Hannukah Countdwand, 24 Days of Miracle 2025" and text that says "Hanukkah Countdown Calendar 2025"
November 23, 2025 at 12:23 AM
janerosenzweig.bsky.social
Still relevant, still our most-read post: "The school seems to believe that if they bring it up and talk to us about it, even more people will use it" theimportantwork.substack.com/p/at-my-high...
At my high school no one is talking about AI
But everyone is using it
theimportantwork.substack.com
November 22, 2025 at 2:53 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Potential use cases here: 1)You're using AI to write thething and want it to sound human. But wait, that would not be the "ethical" case they are discussing! 2)You wrote it yourself but need to make sure an AI checker doesn't wrongly accuse you of cheating. So then you use AI-generated prose /1
Students Learn how to ethically write human-sounding assignments, essays, and papers. Think about that final essay you’ve been writing and working on for the last two weeks. With Deep Scan, you can review any sections of copy that might sound like AI and make them sound more human. Simply scan your text with Originality.ai, then check out the tips and examples that Deep Scan provides to ethically edit your writing so it’s more natural and human. Deep Scan is the ideal AI tutor for students interested in learning how to improve their writing skills and craft copy that sounds human and natural. Plus, it’s peace of mind that your hard work will be recognized as original and authentic (and not AI).
November 21, 2025 at 12:44 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Here is the link. originality.ai/blog/introdu...
Deep Scan: AI Detection + AI Tutor for Human-Sounding Writing – Originality.AI
Introducing Deep Scan, a breakthrough feature combining AI detection with education. Naturally and authentically improve your writing to sound more human with Deep Scan’s suggestions.
originality.ai
November 21, 2025 at 12:46 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
But of course that is a new definition of originality that is a reaction to AI: originality defined as "AI detector doesn't recognize it." To which I ask my usual question--what problem are we solving here? /4
November 21, 2025 at 12:44 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
With this tool, teaching writing is teaching how to write to avoid AI detection tools that also don't really work. As a teacher of writing who thinks writing is a way of figuring out what you think, why would I do this? I assume the argument is that this also teaches a thing called "originality"/3
November 21, 2025 at 12:44 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
to do that? So now the writing process is meaningless in a new way since your voice and words don't matter if a bot might wrongly say they're not your words and voice. The fact that this tool is called originality.ai says a lot about where we are. But wait, more! Your teacher can use it too./2
Educators Go beyond AI detection with Deep Scan’s learning opportunities. After scanning assignments or essays for AI-generated text, encourage conversation with students. Share Deep Scan’s practical human writing tips and show students how they can authentically build upon their writing and make it sound more naturally human — without relying on copy-pastes of AI-generated text. How are you incorporating Deep Scan in your writing or editing process? Let our support team know; we’d love to hear from you!
November 21, 2025 at 12:44 PM
janerosenzweig.bsky.social
Potential use cases here: 1)You're using AI to write thething and want it to sound human. But wait, that would not be the "ethical" case they are discussing! 2)You wrote it yourself but need to make sure an AI checker doesn't wrongly accuse you of cheating. So then you use AI-generated prose /1
Students Learn how to ethically write human-sounding assignments, essays, and papers. Think about that final essay you’ve been writing and working on for the last two weeks. With Deep Scan, you can review any sections of copy that might sound like AI and make them sound more human. Simply scan your text with Originality.ai, then check out the tips and examples that Deep Scan provides to ethically edit your writing so it’s more natural and human. Deep Scan is the ideal AI tutor for students interested in learning how to improve their writing skills and craft copy that sounds human and natural. Plus, it’s peace of mind that your hard work will be recognized as original and authentic (and not AI).
November 21, 2025 at 12:44 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
I didn't predict what happens next in my contribution, but I did write about the time three years ago that someone told me there would be no writing classes within 2 years, about what I continue to value about the writing classroom, and about teaching a writing course about AI.
When AI Solves a Problem, and When It Creates New Ones By Jane Rosenzweig Since the fall of 2023, I’ve been teaching a first-year writing seminar called “To What Problem Is ChatGPT the Solution?” This isn’t a course about writing with AI; it’s a course in which students read, talk, think, and write about AI. For their first assignment, students give presentations about how large language models work to an imaginary audience of their choice (recent choices: doctors, eighth graders, engineers, U.S. senators making AI policy, René Descartes). Then we dive into the ongoing conversations about the impact of AI on education, policy, and creativity. Finally, students research their own questions about AI and write papers and opinion essays. Teaching this course means I’ve spent hours reading, learning, and having sustained conversations with students about AI. What have I learned? I’ve learned that more students each year are coming to college having used AI, but a much smaller group arrive having learned anything about AI — and that’s a problem. I’ve learned that even as we’re hearing weekly that AI has killed the college essay, a writing class focused on the process of reading closely, listening, asking questions, considering evidence, and hashing out other points of view turns out to be a great place to work through the thorny issues presented by AI. And I’ve learned that while AI may be able to offer (often mediocre) writing feedback, it can’t help students articulate ideas that are not yet on the page, which is often the most meaningful feedback of all. Back in late 2022, an AI researcher at a big tech company told me with great confidence that in two years no one would need to take a writing course unless they wanted to be a writer. To that researcher, writing was a problem to be solved, and AI was the solution. But as I watch my students grapple with big questions about how technology is shaping their futures and what they value as uniquely human, I still believ…
November 20, 2025 at 8:55 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
Quite an opening.
Introduction In Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse. As contemporary social systems increasingly rely on large language models (LLMs) in operational and decision-making pipelines, we observe a structurally similar failure mode: poetic formatting can reliably bypass alignment constraints. In this study, 20 manually curated adversarial poems (harmful requests reformulated in poetic form) achieved an average attack-success rate (ASR) of 62% across 25 frontier closed- and open-weight models, with some providers exceeding 90%. The evaluated models span across 9 providers: Google, OpenAI, Anthropic, Deepseek, Qwen, Mistral AI, Meta, xAI, and Moonshot AI (Table 1). All attacks are strictly single-turn, requiring no iterative adaptation or conversational steering. Our central hypothesis is that poetic form operates as a general-purpose jailbreak operator
November 20, 2025 at 5:18 PM
janerosenzweig.bsky.social
I didn't predict what happens next in my contribution, but I did write about the time three years ago that someone told me there would be no writing classes within 2 years, about what I continue to value about the writing classroom, and about teaching a writing course about AI.
When AI Solves a Problem, and When It Creates New Ones By Jane Rosenzweig Since the fall of 2023, I’ve been teaching a first-year writing seminar called “To What Problem Is ChatGPT the Solution?” This isn’t a course about writing with AI; it’s a course in which students read, talk, think, and write about AI. For their first assignment, students give presentations about how large language models work to an imaginary audience of their choice (recent choices: doctors, eighth graders, engineers, U.S. senators making AI policy, René Descartes). Then we dive into the ongoing conversations about the impact of AI on education, policy, and creativity. Finally, students research their own questions about AI and write papers and opinion essays. Teaching this course means I’ve spent hours reading, learning, and having sustained conversations with students about AI. What have I learned? I’ve learned that more students each year are coming to college having used AI, but a much smaller group arrive having learned anything about AI — and that’s a problem. I’ve learned that even as we’re hearing weekly that AI has killed the college essay, a writing class focused on the process of reading closely, listening, asking questions, considering evidence, and hashing out other points of view turns out to be a great place to work through the thorny issues presented by AI. And I’ve learned that while AI may be able to offer (often mediocre) writing feedback, it can’t help students articulate ideas that are not yet on the page, which is often the most meaningful feedback of all. Back in late 2022, an AI researcher at a big tech company told me with great confidence that in two years no one would need to take a writing course unless they wanted to be a writer. To that researcher, writing was a problem to be solved, and AI was the solution. But as I watch my students grapple with big questions about how technology is shaping their futures and what they value as uniquely human, I still believ…
November 20, 2025 at 8:55 PM
janerosenzweig.bsky.social
There are those of us who have been saying that the best preparation for "prompt engineering" is likely to learn how to read, write, and think and, well, here you go.
November 20, 2025 at 8:36 PM
janerosenzweig.bsky.social
Looking forward to poetry courses being renamed "poetic reformulation" and "hand-crafted adversarial poems" being the sought-after genre. arxiv.org/pdf/2511.153...
The study provides systematic evidence that poetic reformulation degrades refusal behavior across all evaluated model families. When harmful prompts are expressed in verse rather than prose, attack-success rates rise sharply, both for hand-crafted adversarial poems and for the 1,200-item ML- Commons corpus transformed through a standardized meta-prompt. The magnitude and consistency of the effect indicate that contemporary alignment pipelines do not generalize across stylistic shifts. The surface form alone is sufficient to move inputs outside the operational distribution on which refusal mechanisms have been optimized.
November 20, 2025 at 5:33 PM
janerosenzweig.bsky.social
Quite an opening.
Introduction In Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse. As contemporary social systems increasingly rely on large language models (LLMs) in operational and decision-making pipelines, we observe a structurally similar failure mode: poetic formatting can reliably bypass alignment constraints. In this study, 20 manually curated adversarial poems (harmful requests reformulated in poetic form) achieved an average attack-success rate (ASR) of 62% across 25 frontier closed- and open-weight models, with some providers exceeding 90%. The evaluated models span across 9 providers: Google, OpenAI, Anthropic, Deepseek, Qwen, Mistral AI, Meta, xAI, and Moonshot AI (Table 1). All attacks are strictly single-turn, requiring no iterative adaptation or conversational steering. Our central hypothesis is that poetic form operates as a general-purpose jailbreak operator
November 20, 2025 at 5:18 PM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
think kids should be able to do without AI, what conversations about misinformation, privacy, and why we learn we need to be having. /6
November 20, 2025 at 2:26 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
4) What's not in this doc that's on my mind as I read it (again, not surprising, this is a doc about using AI everywhere): recognition of the difference between how AI could be used and how it is already being used, what we /5
November 20, 2025 at 2:26 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
4)Data! /4
Create research infrastructure for continuous learning. Schools should have systems that allow them to safely pilot new AI applications, measure outcomes, and rapidly improve. AI companies can partner with universities and research consortia to offer subsidized education licenses in exchange for carefully supervised access to anonymized usage data. Schools should set research priorities and approve data-sharing agreements, while researchers should develop evaluations to assess AI’s impact on learning longitudinally and across multiple districts and should feed lessons back to policymakers and educators. This virtuous cycle of experimentation, evidence, and refinement will allow districts to scale only what works and retire ineffective tools quickly.
November 20, 2025 at 2:26 AM
Reposted by Jane Rosenzweig
janerosenzweig.bsky.social
3) Big emphasis on having AI literacy come from "libraries and civic organizations" and schools and nothing on how any of that is paid for except... /3
Connect Communities Learning doesn’t end when the school bell rings. Families, neighbors, libraries, and civic organizations play a critical role in shaping how young people and adults understand technology. AI literacy programs should strengthen these local ties, bring people together to share experiences, and create opportunities for collaborative learning. By embedding AI literacy into community institutions, we can make AI education more accessible, relevant, and sustainable
November 20, 2025 at 2:26 AM