The quest to build a better AI tutor— from hechingerreport.org by Jill Barshay Researchers make progress with an older ed tech idea: personalized practice
One promising idea has less to do with how an AI tutor explains concepts and more with what it asks students to practice next.
A team at the University of Pennsylvania, which included some AI skeptics, recently tested this approach in a study of close to 800 Taiwanese high school students learning Python programming. All the students used the same AI tutor, which was designed not to give away answers.
But there was one key difference. Half the students were randomly assigned to a fixed sequence of practice problems, progressing from easy to hard. The other half received a personalized sequence with the AI tutor continuously adjusting the difficulty of each problem based on how the student was performing and interacting with the chatbot.
The idea is based on what educators call the “zone of proximal development.” When problems are too easy, students get bored. When they’re too hard, students get frustrated. The goal is to keep students in a sweet spot: challenged, but not overwhelmed.
The researchers found that students in the personalized group did better on a final exam than students in the fixed problem group. The difference was characterized as the equivalent of 6 to 9 months of additional schooling, an eye-catching claim for an after-school online course that lasted only five months. … To address this, Chung’s team combined a large language model with a separate machine-learning algorithm that analyzes how students interact with the online course platform — how they answer the practice questions, how many times they revise or edit their coding, and the quality of their conversations with the chatbot — and uses that information to decide which problem to serve up next.
The real story isn’t what AI can produce — it’s how it changes the decisions we make at every stage of instructional design.
After working with thousands of instructional designers on my bootcamp, I’ve learned something counterintuitive: the best teams aren’t the ones with the fanciest AI tools — they’re the ones who know when to use which mode—and when to use none at all.
Once you recognise that, you start to see instructional design differently — not as a linear process, but as a series of decision loops where AI plays distinct roles.
In this post, I show you the 3 modes of AI that actually matter in instructional design — and map them across every phase of ADDIE so you know exactly when to let AI run, and when to slow down and think.
In higher education, developing strong multiple-choice questions can be a time-intensive part of the course design process. Developing such items requires subject-matter expertise and assessment literacy, and for faculty and designers who are creating and producing online courses, it can be difficult to find the capacity to craft quality multiple-choice questions.
At the University of Michigan Center for Academic Innovation, learning experience designers are using generative artificial intelligence to streamline the multiple-choice question development process and help ameliorate this issue. In this article, I summarize one of our projects that explored effective prompting strategies to develop multiple-choice questions with ChatGPT for our open course portfolio. We examined how structured prompting can improve the quality of AI-generated assessments, producing relevant comprehension and recall items and options that include plausible distractors.
Achieving this goal enables us to develop several ungraded practice opportunities, preparing learners for their graded assessments while also freeing up more time for course instructors and designers.
Highly Complementary Capabilities Will Create a Leading Technology Platform, Redefining Skills Discovery, Development, and Mastery for Learners and Organizations at Scale
Unites Udemy’s Dynamic AI-Powered Skills Development Marketplace with World-Class University and Industry Brands Under the Coursera Ecosystem, Expanding Value, Impact, and Choice Globally
Strengthens Combined Company’s Financial Profile with Pro Forma Annual Revenue of More Than $1.5 Billion and Anticipated Annual Run-Rate Cost Synergies of $115 Million Within 24 Months
“We’re at a pivotal moment in which AI is rapidly redefining the skills required for every job across every industry. Organizations and individuals around the world need a platform that is as agile as the new and emerging skills learners must master,” said Greg Hart, CEO of Coursera. “By combining the highly complementary strengths of Coursera and Udemy, we will be in an even stronger position to address the global talent transformation opportunity, unlock a faster pace of innovation, and deliver valuable experiences and outcomes for our learners and customers. Together, we will ensure our millions of learners, thousands of enterprise, university, and government customers, and expert instructors have a platform to keep pace with technology acceleration.”
You’ll hear me briefly describe five recent op-eds on teaching and learning in higher ed. For each op-ed, I’ll ask each of our panelists if they “take it,” that is, generally agree with the main thesis of the essay, or “leave it.” This is an artificial binary that I’ve found to generate rich discussion of the issues at hand.
I just completed nearly 60,000 miles of travel across Europe, Asia, and the Middle East meeting with hundred of companies to discuss their AI strategies. While every company’s maturity is different, one thing is clear: AI as a business tool has arrived: it’s real and the use-cases are growing.
A new survey by Wharton shows that 46% of business leaders use Gen AI daily and 80% use it weekly. And among these users, 72% are measuring ROI and 74% report a positive return. HR, by the way, is the #3 department in use cases, only slightly behind IT and Finance.
What are companies getting out of all this? Productivity. The #1 use case, by far, is what we call “stage 1” usage – individual productivity.
.
From DSC: Josh writes: “Many of our large clients are now implementing AI-native learning systems and seeing 30-40% reduction in staff with vast improvements in workforce enablement.”
While I get the appeal (and ROI) from management’s and shareholders’ perspective, this represents a growing concern for employment and people’s ability to earn a living.
And while I highly respect Josh and his work through the years, I disagree that we’re over the problems with AI and how people are using it:
Two years ago the NYT was trying to frighten us with stories of AI acting as a romance partner. Well those stories are over, and thanks to a $Trillion (literally) of capital investment in infrastructure, engineering, and power plants, this stuff is reasonably safe.
Those stories are just beginning…they’re not close to being over.
So let’s imagine a world where there’s no separation between learning and assessment: it’s ongoing. There’s always assessment, always learning, and they’re tied together. Then we can ask: what is the role of the human in that world? What is it that AI can’t do?
…
Imagine something like that in higher ed. There could be tutoring or skill-based work happening outside of class, and then relationship-based work happening inside of class, whether online, in person, or some hybrid mix.
The aspects of learning that don’t require relational context could be handled by AI, while the human parts remain intact. For example, I teach strategy and strategic management. I teach people how to talk with one another about the operation and function of a business. I can help students learn to be open to new ideas, recognize when someone pushes back out of fear of losing power, or draw from my own experience in leading a business and making future-oriented decisions.
But the technical parts such as the frameworks like SWOT analysis, the mechanics of comparing alternative viewpoints in a boardroom—those could be managed through simulations or reports that receive immediate feedback from AI. The relational aspects, the human mentoring, would still happen with me as their instructor.
My take is this: in all of the anxiety lies a crucial and long-overdue opportunity to deliver better learning experiences. Precisely because Atlas perceives the same context in the same moment as you, it can transform learning into a process aligned with core neuro-scientific principles—including active retrieval, guided attention, adaptive feedback and context-dependent memory formation.
Perhaps in Atlas we have a browser that for the first time isn’t just a portal to information, but one which can become a co-participant in active cognitive engagement—enabling iterative practice, reflective thinking, and real-time scaffolding as you move through challenges and ideas online.
With this in mind, I put together 10 use cases for Atlas for you to try for yourself.
…
6. Retrieval Practice
What: Pulling information from memory drives retention better than re-reading. Why: Practice testing delivers medium-to-large effects (Adesope et al., 2017). Try: Open a document with your previous notes. Ask Atlas for a mixed activity set: “Quiz me on the Krebs cycle—give me a near-miss, high-stretch MCQ, then a fill-in-the-blank, then ask me to explain it to a teen.” Atlas uses its browser memory to generate targeted questions from your actual study materials, supporting spaced, varied retrieval.
From DSC: A quick comment. I appreciate these ideas and approaches from Katarzyna and Rita. I do think that someone is going to want to be sure that the AI models/platforms/tools are given up-to-date information and updated instructions — i.e., any new procedures, steps to take, etc. Perhaps I’m missing the boat here, but an internal AI platform is going to need to have access to up-to-date information and instructions.
In that spirit, in this post I examine a report from Virginia’s Joint Legislative Audit and Review Commission (JLARC) on Virginia’s Community Colleges and the changing higher-education landscape. The report offers a rich view of how several major issues are evolving at the institutional level over time, an instructive case study in big changes and their implications.
Its empirical depth also prompts broader questions we should ask across higher education.
What does the shift toward career education and short-term training mean for institutional costs and funding?
How do we deliver effective student supports as enrollment moves online?
As demand shifts away from on-campus learning, do physical campuses need to get smaller?
Are we seeing a generalizable movement from academic programs to CTE to short-term options? If so, what does that imply for how community colleges are staffed and funded?
As online learning becomes a larger, permanent share of enrollment, do student services need a true bimodal redesign, built to serve both online and on-campus students effectively? Evidence suggests this urgent question is not being addressed, especially in cash-strapped community colleges.
As online learning grows, what happens to physical campuses? Improving space utilization likely means downsizing, which carries other implications. Campuses are community anchors, even for online students—so finding the right balance deserves serious debate.
That gap creates compliance risk and wasted investment. It leaves HR leaders with a critical question: How do you measure and validate real learning when AI is doing the work for employees?
Designing Training That AI Can’t Fake
Employees often find static slide decks and multiple-choice quizzes tedious, while AI can breeze through them. If employees would rather let AI take training for them, it’s a red flag about the content itself.
One of the biggest risks with agentic AI is disengagement. When AI can complete a task for employees, their incentive to engage disappears unless they understand why the skill matters, Rashid explains. Personalization and context are critical. Training should clearly connect to what employees value most – career mobility, advancement, and staying relevant in a fast-changing market.
Nearly half of executives believe today’s skills will expire within two years, making continuous learning essential for job security and growth. To make training engaging, Rashid recommends:
Delivering content in formats employees already consume – short videos, mobile-first modules, interactive simulations, or micro-podcasts that fit naturally into workflows. For frontline workers, this might mean replacing traditional desktop training with mobile content that integrates into their workday.
Aligning learning with tangible outcomes, like career opportunities or new responsibilities.
Layering in recognition, such as digital badges, leaderboards, or team shout-outs, to reinforce motivation and progress
Microsoft is pitching a recent shift of AI agents in Microsoft Teams as more than just smarter assistance. Instead, these agents are built to behave like human teammates inside familiar apps such as Teams, SharePoint, and Viva Engage. They can set up meeting agendas, keep files in order, and even step in to guide community discussions when things drift off track.
…
Unlike tools such as ChatGPT or Claude, which mostly wait for prompts, Microsoft’s agents are designed to take initiative. They can chase up unfinished work, highlight items that still need decisions, and keep projects moving forward. By drawing on Microsoft Graph, they also bring in the right files, past decisions, and context to make their suggestions more useful.
As an advisor to Aibrary, I am impressed with their educational philosophy, which is based both on theory and on empirical research findings. Aibrary is an innovative approach to self-directed learning that complements academic resources. Expanding our historic conceptions of books, libraries, and lifelong learning to new models enabled by emerging technologies is central to empowering all of us to shape our future. .
Why AI literacy must come before policy — from timeshighereducation.com by Kathryn MacCallum and David Parsons When developing rules and guidelines around the uses of artificial intelligence, the first question to ask is whether the university policymakers and staff responsible for implementing them truly understand how learners can meet the expectations they set
Literacy first, guidelines second, policy third
For students to respond appropriately to policies, they need to be given supportive guidelines that enact these policies. Further, to apply these guidelines, they need a level of AI literacy that gives them the knowledge, skills and understanding required to support responsible use of AI. Therefore, if we want AI to enhance education rather than undermine it, we must build literacy first, then create supportive guidelines. Good policy can then follow.
Sept 22 (Reuters) – At orientation last month, 375 new Fordham Law students were handed two summaries of rapper Drake’s defamation lawsuit against his rival Kendrick Lamar’s record label — one written by a law professor, the other by ChatGPT.
The students guessed which was which, then dissected the artificial intelligence chatbot’s version for accuracy and nuance, finding that it included some irrelevant facts.
The exercise was part of the first-ever AI session for incoming students at the Manhattan law school, one of at least eight law schools now incorporating AI training for first-year students in orientation, legal research and writing courses, or through mandatory standalone classes.
In this episode, we explore why digital accessibility can be so important to the student experience. My guest is Amy Lomellini, director of accessibility at Anthology, the company that makes the learning management system Blackboard. Amy teaches educational technology as an adjunct at Boise State University, and she facilitates courses on digital accessibility for the Online Learning Consortium. In our conversation, we talk about the importance of digital accessibility to students, moving away from the traditional disclosure-accommodation paradigm, AI as an assistive technology, and lots more.
A day in the life: The next 25 years A learner wakes up. Their AI-powered learning coach welcomes them, drawing their attention to their progress and helping them structure their approach to the day. A notification reminds them of an upcoming interview and suggests reflections to add to their learning portfolio.
Rather than a static gradebook, their portfolio is a dynamic, living record, curated by the student, validated by mentors in both industry and education, and enriched through co-creation with maturing modes of AI. It tells a story through essays, code, music, prototypes, journal reflections, and team collaborations. These artifacts are not “submitted”, they are published, shared, and linked to verifiable learning outcomes.
And when it’s time to move, to a new institution, a new job, or a new goal, their data goes with them, immutable, portable, verifiable, and meaningful.
From DSC: And I would add to that last solid sentence that the learner/student/employee will be able to control who can access this information. Anyway, some solid reflections here from Lev.
I know a lot of readers will disagree with this, and the timeline feels aggressive (the future always arrives more slowly than pundits expect) but I think the overall premise is sound: “The concept of a tipping point in education – where AI surpasses traditional schools as the dominant learning medium – is increasingly plausible based on current trends, technological advancements, and expert analyses.”
The Rundown: In this tutorial, you will learn how to combine NotebookLM with ChatGPT to master any subject faster, turning dense PDFs into interactive study materials with summaries, quizzes, and video explanations.
Step-by-step:
Go to notebooklm.google.com, click the “+” button, and upload your PDF study material (works best with textbooks or technical documents)
Choose your output mode: Summary for a quick overview, Mind Map for visual connections, or Video Overview for a podcast-style explainer with visuals
Generate a Study Guide under Reports — get Q&A sets, short-answer questions, essay prompts, and glossaries of key terms automatically
Take your PDF to ChatGPT and prompt: “Read this chapter by chapter and highlight confusing parts” or “Quiz me on the most important concepts”
Combine both tools: Use NotebookLM for quick context and interactive guides, then ChatGPT to clarify tricky parts and go deeperPro Tip: If your source is in EPUB or audiobook, convert it to PDF before uploading. Both NotebookLM and ChatGPT handle PDFs best.
Claude can now create and edit Excel spreadsheets, documents, PowerPoint slide decks, and PDFs directly in Claude.ai and the desktop app. This transforms how you work with Claude—instead of only receiving text responses or in-app artifacts, you can describe what you need, upload relevant data, and get ready-to-use files in return.
Also see:
Microsoft to lessen reliance on OpenAI by buying AI from rival Anthropic — from techcrunch.com byRebecca Bellan
Microsoft will pay to use Anthropic’s AI in Office 365 apps, The Information reports, citing two sources. The move means that Anthropic’s tech will help power new features in Word, Excel, Outlook, and PowerPoint alongside OpenAI’s, marking the end of Microsoft’s previous reliance solely on the ChatGPT maker for its productivity suite. Microsoft’s move to diversify its AI partnerships comes amid a growing rift with OpenAI, which has pursued its own infrastructure projects as well as a potential LinkedIn competitor.
In this episode of Unfixed, we talk with Ray Schroeder—Senior Fellow at UPCEA and Professor Emeritus at the University of Illinois Springfield—about Artificial General Intelligence (AGI) and what it means for the future of higher education. While most of academia is still grappling with ChatGPT and basic AI tools, Schroeder is thinking ahead to AI agents, human displacement, and AGI’s existential implications for teaching, learning, and the university itself. We explore why AGI is so controversial, what institutions should be doing now to prepare, and how we can respond responsibly—even while we’re already overwhelmed.
Data from the State of AI and Instructional Design Report revealed that 95.3% of the instructional designers interviewed use AI in their daily work [1]. And over 85% of this AI use occurs during the design and development process.
These figures showcase the immense impact AI is already having on the instructional design world.
If you’re an L&D professional still on the fence about adding AI to your workflow or an AI convert looking for the next best tools, keep reading.
This guide breaks down 5 of the top AI tools for instructional designers in 2025, so you can streamline your development processes and build better training faster.
But before we dive into the tools of the trade, let’s address the elephant in the room:
Miro and GenAI as drivers of online student engagement — from timeshighereducation.com by Jaime Eduardo Moncada Garibay A set of practical strategies for transforming passive online student participation into visible, measurable and purposeful engagement through the use of Miro, enhanced by GenAI
To address this challenge, I shifted my focus from requesting participation to designing it. This strategic change led me to integrate Miro, a visual digital workspace, into my classes. Miro enables real-time visualisation and co-creation of ideas, whether individually or in teams.
…
The transition from passive attendance to active engagement in online classes requires deliberate instructional design. Tools such as Miro, enhanced by GenAI, enable educators to create structured, visually rich learning environments in which participation is both expected and documented.
While technology provides templates, frames, timers and voting features, its real pedagogical value emerges through intentional facilitation, where the educator’s role shifts from delivering content to orchestrating collaborative, purposeful learning experiences.
In the past, it was typical for faculty to teach online courses as an “overload” of some kind, but BOnES data show that 92% of online programs feature courses taught as part of faculty member’s standard teaching responsibilities. Online teaching has become one of multiple modalities in which faculty teach regularly.
Three-quarters of chief online officers surveyed said they plan to have a great market share of online enrollments in the future, but only 23% said their current marketing is better than their competitors. The rising tide of online enrollments won’t lift all boats–some institutions will fare better than others.
Staffing at online education units is growing, with the median staff size increasing from 15 last year to 20 this year. Julie pointed out that successful online education requires investment of resources. You might need as many buildings as onsite education does, but you need people and you need technology.
The Online Education Marketplace Is Increasingly Competitive: …
Alternative Credentials Take Center Stage: …
AI Integration Lacks Strategic Coordination: …
Just 28% of faculty are considered fully prepared for online course design, and 45% for teaching. Alarmingly, only 28% of institutions report having fully developed academic continuity plans for future emergency pivots to online.
Cultural resistance remains strong. Many [Chief Online Learning Officers] COLOs say faculty and deans still believe in-person learning is “just better,” creating headwinds even for modest online growth. As one respondent at a four-year institution with a large online presence put it:
Supportive departments [that] see the value in online may have very different levels of responsiveness compared to academic departments [that] are begrudgingly online. There is definitely a growing belief that students “should” be on-ground and are only choosing online because it’s easy/ convenient. Never mind the very real and growing population of nontraditional learners who can only take online classes, and the very real and growing population of traditional-aged learners who prefer online classes; many faculty/deans take a paternalistic, “we know what’s best” approach.
… Ultimately, what we need is not just more ambition but better ambition. Ambition rooted in a realistic understanding of institutional capacity, a shared strategic vision, investments in policy and infrastructure, and a culture that supports online learning as a core part of the academic mission, not an auxiliary one. It’s time we talked about what it really takes to grow online learning , and where ambition needs to be matched by structure.
From DSC: Yup. Culture is at the breakfast table again…boy, those strategies taste good.
I’d like to take some of this report — like the graphic below — and share it with former faculty members and members of a couple of my past job families’ leadership. They strongly didn’t agree with us when we tried to advocate for the development of online-based learning/programs at our organizations…but we were right. We were right all along. And we were LEADING all along. No doubt about it — even if the leadership at the time said that we weren’t leading.
The cultures of those organizations hurt us at the time. But our cultivating work eventually led to the development of online programs — unfortunately, after our groups were disbanded, they had to outsource those programs to OPMs.
Arizona State University — with its dramatic growth in online-based enrollments.
In this post, part of the UsableNet 25th anniversary series, I’m taking a look at where things stand in 2025. I’ll discuss the areas that have improved—such as online shopping, banking, and social media—and the ones that still make it challenging to perform basic tasks, including travel, healthcare, and mobile apps. I hope that by sharing what works and what doesn’t, I can help paint a clearer picture of the digital world as it stands today.
On June 28, 2025, the European Accessibility Act (EAA) officially became enforceable across the European Union. This law requires digital products and services—including websites, mobile apps, e-commerce platforms, and software to meet the defined accessibility standards outlined in EN 301 549, which aligns with the WCAG 2.1 Level AA.
Companies that serve EU consumers must be able to demonstrate that accessibility is built into the design, development, testing, and maintenance of their digital products and services.
This milestone also arrives as UsableNet celebrates 25 years of accessibility leadership—a moment to reflect on how far we’ve come and what digital teams must do next.
Intellectual rigor comes from the journey: the dead ends, the uncertainty, and the internal debate. Skip that, and you might still get the insight–but you’ll have lost the infrastructure for meaningful understanding. Learning by reading LLM output is cheap. Real exercise for your mind comes from building the output yourself.
The irony is that I now know more than I ever would have before AI. But I feel slightly dumber. A bit more dull. LLMs give me finished thoughts, polished and convincing, but none of the intellectual growth that comes from developing them myself.
Every few months I put together a guide on which AI system to use. Since I last wrote my guide, however, there has been a subtle but important shift in how the major AI products work. Increasingly, it isn’t about the best model, it is about the best overall system for most people. The good news is that picking an AI is easier than ever and you have three excellent choices. The challenge is that these systems are getting really complex to understand. I am going to try and help a bit with both.
First, the easy stuff.
Which AI to Use For most people who want to use AI seriously, you should pick one of three systems: Claude from Anthropic, Google’s Gemini, and OpenAI’s ChatGPT.
This summer, I tried something new in my fully online, asynchronous college writing course. These classes have no Zoom sessions. No in-person check-ins. Just students, Canvas, and a lot of thoughtful design behind the scenes.
One activity I created was called QuoteWeaver—a PlayLab bot that helps students do more than just insert a quote into their writing.
It’s a structured, reflective activity that mimics something closer to an in-person 1:1 conference or a small group quote workshop—but in an asynchronous format, available anytime. In other words, it’s using AI not to speed students up, but to slow them down.
…
The bot begins with a single quote that the student has found through their own research. From there, it acts like a patient writing coach, asking open-ended, Socratic questions such as:
What made this quote stand out to you?
How would you explain it in your own words?
What assumptions or values does the author seem to hold?
How does this quote deepen your understanding of your topic?
It doesn’t move on too quickly. In fact, it often rephrases and repeats, nudging the student to go a layer deeper.
On [6/13/25], UNESCO published a piece I co-authored with Victoria Livingstone at Johns Hopkins University Press. It’s called The Disappearance of the Unclear Question, and it’s part of the ongoing UNESCO Education Futures series – an initiative I appreciate for its thoughtfulness and depth on questions of generative AI and the future of learning.
Our piece raises a small but important red flag. Generative AI is changing how students approach academic questions, and one unexpected side effect is that unclear questions – for centuries a trademark of deep thinking – may be beginning to disappear. Not because they lack value, but because they don’t always work well with generative AI. Quietly and unintentionally, students (and teachers) may find themselves gradually avoiding them altogether.
Of course, that would be a mistake.
We’re not arguing against using generative AI in education. Quite the opposite. But we do propose that higher education needs a two-phase mindset when working with this technology: one that recognizes what AI is good at, and one that insists on preserving the ambiguity and friction that learning actually requires to be successful.
By leveraging generative artificial intelligence to convert lengthy instructional videos into micro-lectures, educators can enhance efficiency while delivering more engaging and personalized learning experiences.
Researchers at Massachusetts Institute of Technology (MIT) have now devised a way for LLMs to keep improving by tweaking their own parameters in response to useful new information.
The work is a step toward building artificial intelligence models that learn continually—a long-standing goal of the field and something that will be crucial if machines are to ever more faithfully mimic human intelligence. In the meantime, it could give us chatbots and other AI tools that are better able to incorporate new information including a user’s interests and preferences.
The MIT scheme, called Self Adapting Language Models (SEAL), involves having an LLM learn to generate its own synthetic training data and update procedure based on the input it receives.
Edu-Snippets — from scienceoflearning.substack.com by Nidhi Sachdeva and Jim Hewitt Why knowledge matters in the age of AI; What happens to learners’ neural activity with prolonged use of LLMs for writing
Highlights:
Offloading knowledge to Artificial Intelligence (AI) weakens memory, disrupts memory formation, and erodes the deep thinking our brains need to learn.
Prolonged use of ChatGPT in writing lowers neural engagement, impairs memory recall, and accumulates cognitive debt that isn’t easily reversed.
