About once a month, a student or colleague brings up the MIT critical thinking paper. Once a month I send them to the wonderful @cathicks.bsky.social episode on it. We have hunches about AI and learning that will likely be proven correct. We don't need flawed studies to do that. Link in comments.

You deserve better brain research - Change, Technically

SHOW NOTES:For an example of a consideration of learning with information searching, a paper by Saskia Giebl and co-authors explored students learning basic programming concepts aided with a search en...

changetechnically.fyi

What's better for learning: Working through challenging concepts or evaluating flawed AI outputs? Colleague Ethan Dickey ran a randomized study of 220 students. They showed no difference in skill gain for any particular method. arxiv.org/abs/2607.27586

Is Solving Better Than Evaluating GenAI Solutions?

As Generative AI (GenAI) tools become increasingly capable of generating solutions to computing assignments, the computing education community is exploring pedagogical approaches that emphasize soluti...

arxiv.org

How my colleague, Dan Guberman, has been using collaborative grading to markedly increase motivation and student engagement in an asynchronous, online course. www.chronicle.com/newsletter/t...

How Collaborative Grading Is Designed to Motivate Students

One instructor found a different approach that puts the emphasis on setting goals students want to achieve.

chronicle.com

Someone hawking face recognition quiz software for teachers on professional listserv. Pushed Gemini to make one. 15 minutes later, I have a dynamic scoring app. It quizzes me to match names to faces and a name to many faces, with facial recognition to group like with like and all run locally.

This is awful. I cite Jane’s fantastic Boston Globe column on AI writing tone all the time. She is a leader in student writing learning gains and a wonderful author herself.

Mary Lewis@profmarylewis.bsky.social · last wk.

It‘s quite possible that with this move, @harvard.edu could become the only Ivy League university without a writing center. www.thecrimson.com/article/2026...

It‘s quite possible that with this move, @harvard.edu could become the only Ivy League university without a writing center. www.thecrimson.com/article/2026...

Harvard College Lays Off Writing Center Director, Consolidates Center | News | The Harvard Crimson

The Harvard College Writing Center will no longer operate as a standalone unit after its director was laid off and its functions were consolidated as part of a restructuring in the Faculty of Arts and...

thecrimson.com

Again, we all had the hunch, but love how this demonstrates that the act of tutoring can mitigate AI harm to cognitive practice. Trick is, how to get the students to use a modified, limited software? www.pnas.org/doi/10.1073/...

Generative AI without guardrails can harm learning: Evidence from high school mathematics | PNAS

Generative AI is poised to revolutionize how humans work, and has already demonstrated promise in significantly improving human productivity. A key...

pnas.org

Lot of AI+learning hunches getting empirical over the last three years. Like this look at student metacognition with AI in traditional vs experiential learning environments. Gives further credo for relevance to me/my future profession more directly in coursework. arxiv.org/abs/2607.20047

Experiential Versus Instructional Approaches for Eliciting Metacognitive Awareness in AI-Assisted Learning: A Short-Term Longitudinal Study

With generative AI (GenAI) entering classrooms the question to which teaching approach best supports metacognitive skill acquisition in AI-assisted learning becomes pressing. In this short-term longit...

arxiv.org

I quibble with the "stakes" in this experiment, but general results are marked. AI access makes people less likely to respond to a question with "I don't know" even as their answers are less accurate. Side thought - How can we quantify stakes in coursework??https://arxiv.org/abs/2607.13562

I took the convo about hamburger helper and I researched it lol open.substack.com/pub/kyleepen...

AI’s Hamburger Helper Test

What a delicious convenience meal taught me about the honest water math of AI

open.substack.com

Has anyone tried casual cursing at AI cowork? Like, both in your initial prompting and critiquing outputs?

I'm REALLY not one to brag, but Claude has been telling me that I am making lots of good catches lately.

Between old gaming GPUs and new "cheap" unified-memory boxes, inference in the home is increasingly feasible. Not frontier, but real. Thing is, when you have free inference you tend to use inference for stuff that really, honestly, doesn't require inference. Today I gave our cat a lawyer.

@billycarpy.bsky.social is it wrong to actively root against England, Ecuador, Sweden, Germany, France, Norway and Brazil? Asking for a friend

Student AI-use by Bloom's Dimension arxiv.org/abs/2606.29442. This seems about right. Imagine most students thinking: "Explain this thing" or "How do I do this thing." All normal caveats of student honesty through monitored systems, but more of this, please.

LLMs are not great at revising abstracts, overall paper rhetoric/global tone. Humans are also not great at fixing issues with large AI-generated text. Local sentence structure seems the only consistent value-add.

Process-Oriented Evaluation of AI-Assisted Scientific Writing

Bad writing hinders the publication of science. The role of artificial intelligence (AI) in generating and editing scientific texts remains unsettled. Abstracts serve as the critical gateway to scient...

arxiv.org

There is a particular feeling of connectedness when a singular thing in my Venn diagram of life interests and experiences aligns perfectly with EXACTLY two of my friends'. One feels nice, three feels almost like pandering.

Increasingly convinced that imposition of AI via centralized "tutors" and homogenized tools will remain as burdensome as most LMS software. Ariel Stilerman at Stanford provides my favorite example of localized, disciplinary de-programming/de-training LLM structures. arxiv.org/abs/2606.01410

What LLMs Must Forget to Teach Effectively: A DIY Approach to Premodern Japanese Language Pedagogy

We discuss a novel approach to Premodern Japanese Language Pedagogy (PJLP) with potential applications in other languages and fields. The integration of artificial intelligence into education has larg...

arxiv.org

First attempt I've seen to gauge the frequency of sycophancy in AI "tutoring" uses. Authors conclude ~14% sycophancy rates: Chat is susceptible to authoritative pressure, Claude to context-switching. arxiv.org/abs/2605.14604

Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks

This position paper argues that effective tutoring requires corrective friction: surfacing misconceptions and challenging them supportively to drive conceptual change. Yet preference-aligned LLMs can ...

arxiv.org

Need backup copies of every document you give to Agentic AI, I suppose: arxiv.org/abs/2604.15597

LLMs Corrupt Your Documents When You Delegate

Large Language Models (LLMs) are poised to disrupt knowledge work, with the emergence of delegated work as a new interaction paradigm (e.g., vibe coding). Delegation requires trust - the expectation t...

arxiv.org

One of the student groups in my AI class created an awesome student guide using Prezi and it took all my appreciation of their work to get through that interface. It was worth it. Do students use/value Prezi?

Pretty sure that almost all technologies fall prey to this: Without lots of guidance, students will use AI only near deadlines and only for quick answers despite any conversational tutor guardrails and global prompts. arxiv.org/pdf/2604.23486 Will students ever use suggested follow-up prompts?

I use AI a lot in educational research and it holds great promise for data categorization and analysis. But not everything in education needs to be analyzed. Case in point: Student-teacher dialogue. Interesting technical achievement but I keep asking "why"?? arxiv.org/abs/2604.21137

Enhancing Science Classroom Discourse Analysis through Joint Multi-Task Learning for Reasoning-Component Classification

Analyzing the reasoning patterns of students in science classrooms is critical for understanding knowledge construction mechanism and improving instructional practice to maximize cognitive engagement,...

arxiv.org

Dispatch from my demographic! Working moms have Prompt Parties to swap prompts to automate/outsource mental load. We're served content about how to make ChatGPT a house manager, virtual assistant, etc. Women's uptake isn't slightly surprising, because nothing else has fixed imbalanced mental load.

Stefan Schubert@stefanschubert.bsky.social · 4mo ago

When ChatGPT was launched, almost all users were men – but now, women are the majority www.update.news/p/will-ai-dr...

Has anyone started to write their professional stuff with an eye towards how an LLM-guided lit review would evaluate it?

@yahtzeecroshaw.bsky.social Hiya Ben -- Hi Ben, Freddie Wong and I work on a comedy podcast called Dungeons and Daddies and we were wondering if you'd like to make a brief pre-recorded cameo at the top of our next episode.

Of the many, many daily disappointments that accompany modern academic technological tools, none have evoked as many scowls from me as @developers.google.com 's Voice typing function in Docs.

As with inauthentic assignments, when we approach teaching evaluations as time-sensitive menial tasks in an effort to quantify the learning process, both students and instructors will find ways to obviate the activity. If you care about evals, consider conducting your own formative ones.

Marc Watkins@marcwatkins.bsky.social · 4mo ago

We've only begun to think about the risks AI agents pose to faculty labor. Grading is one area. Consider if a student uses an agent to fill out a teacher evaluation and the upstream consequences of bias impacting promotion, tenure, or renewal. open.substack.com/pub/marcwatk...

We've only begun to think about the risks AI agents pose to faculty labor. Grading is one area. Consider if a student uses an agent to fill out a teacher evaluation and the upstream consequences of bias impacting promotion, tenure, or renewal. open.substack.com/pub/marcwatk...

Who (Or What) Filled Out Your Course Evaluation?

Canvas will introduce its first AI agent directly into the LMS later this year, arriving in a chaotic landscape where few institutions have considered how agentic AI will impact faculty teaching and s...

open.substack.com

In 1993, I got my first taste of the internet through IRC (Internet Relay Chat). It was boggling how typed lines of input would nearly instantly receive a nuanced, contextual response. Shifted my paradigm on keyboard input. I've had the same feeling after a few months with Claude Code.