Georgia Tech and edX test AI-powered course on AI

Today we launched our first course using an AI Avatar with Georgia Institute of Technology's David Joyner, who is leveraging his AI avatar ("DAI-vid") to co-teach his course Foundations of Generative AI on edX. This is an excellent example of how technology can amplify great teaching - responsible innovation that keeps learning personal, scalable, and world-class. Read more below 👇

Innovation in education is most powerful when it strengthens—not replaces—the human connection. Today, Georgia Institute of Technology launched Foundations of Generative AI on edX—a new course that uses an AI avatar of Professor David Joyner to teach about the very technology that built it. It’s a thoughtful, responsible experiment in how AI can help educators keep courses current, scale their reach, and focus more on what they do best: inspiring learners. At 2U, we’re proud to support partners like Georgia Tech who are advancing the future of learning with purpose. Learn more → https://lnkd.in/ex5d67ea

What happens when you use AI to teach AI? At edX, we teamed up with David Joyner from Georgia Tech to find out. Together we launched Foundations of Generative AI—one of the first open courses built with AI itself. David wrote and designed the content, while tools like HeyGen, GradyAI, and Xpert brought it to life with AI-generated lectures, grading, and learner support. This pilot shows what’s possible when we combine AI + HI (Artificial + Human Intelligence)—AI handles the routine, while educators focus on creativity, mentoring, and connection. As edX founder Anant Agarwal said, “AI can give every educator their own ‘super-TA.’” The course is free, self-paced, and offers a first-hand look at how generative AI actually works—while experiencing it in action. #GenerativeAI #AIinEducation #TeachingWithAI #FutureOfLearning #edX

A publication which cuts through the noise on AI in education with a simple, logical structure; problem, opportunity, action. It doesn’t overhype the use of technology but shows exactly how AI can make learning richer and assessments more useful. 'no matter how powerful AI systems become, they will only ever be part of a learning process: components within a larger system whose ultimate measures of failure or success remain wholly human.' AI doesn't have to be a shortcut, it can be a catalyst for deeper learning and educators must play a central role in shaping it. https://lnkd.in/ew7tNz34

Carl Hendrick’s recent piece raises an important question about the future of education. As algorithms take on more of the work of remembering, organising, and even thinking for us, we cannot assume that convenience leads to understanding. If learning becomes too frictionless, we risk losing the slow and effortful processes that help students form deep knowledge. So the challenge ahead is not just adopting technology, but ensuring our tools still allow learners to think, struggle a little, and make meaning for themselves. The future of education is not about replacing thinking, but protecting it. #FutureOfEducation #LearningDesign #AI #EdTech https://lnkd.in/gJ7n7YPY

Learning cannot be outsourced to #AI – This is how AI is transforming university education Those teaching at the #University must consider both the pros and cons of #artificialintelligence. Training students in the responsible use of AI is important, but learning still requires traditional #reading and #writing skills. These cannot be outsourced to generative AI without harm to learning! https://lnkd.in/d8buDmRW

The Best Way to Use AI for Learning Many people use AI to boost learning efficiency. For example, by summarizing dense material into concise bullet points or translating foreign texts into their native language. Essentially this allows people to do what they were already doing, only faster. But these efficiency gains, when accumulated, reveal a deeper transformation: learning methods that once felt too time-consuming to attempt have suddenly become feasible. As a result, the most important question to ask is not “how to learn more efficiently,” but “how to become capable of learning knowledge that is more complex, abstract, and challenging.” https://lnkd.in/dQMaB87i #TrustEverybodyButCutTheCards

🧩 Nested Learning: The Illusion of Deep Learning Architectures What if “depth” in Deep Learning isn’t about stacking layers — but stacking learning loops? A new paper, “Nested Learning: The Illusion of Deep Learning Architectures” (arXiv:2508.19597), challenges one of AI’s biggest assumptions — that making models deeper automatically makes them smarter. 🔁 A Paradigm Shift Traditional deep learning builds taller networks. Nested Learning (NL), however, builds smarter learners — systems that learn at multiple speeds and depths of understanding. Instead of a single global optimization, NL creates layers of learning inside learning — fast inner loops adjusting local behavior and slower outer loops refining long-term strategy. Think of it like this: The neurons don’t just learn — they learn how to learn at different rhythms. ⚙️ Why It Matters 🔹 Beyond Illusion of Depth – Real intelligence might come from how updates propagate across time, not layers. 🔹 Resilient Memory – Nested models naturally support continual learning without catastrophic forgetting. 🔹 Adaptive Systems – Models can evolve organically, with each “nested layer” adjusting to a unique learning horizon. This approach reframes the architecture question entirely — depth ≠ complexity, adaptivity = intelligence. 🧠 The Takeaway Nested Learning is more than an algorithmic tweak — it’s a mindset shift. It asks: “Can we design AI systems that grow wiser, not just wider?” By embedding loops within loops, AI could soon achieve a kind of self-reflective cognition — where learning never truly stops, just changes tempo. 📄 Paper: Nested Learning: The Illusion of Deep Learning Architectures (arXiv): https://lnkd.in/gjRxhutQ #NestedLearning #DeepLearning #AIResearch #AdaptiveAI #MachineLearning #AgenticAI #NeuralNetworks #LLMs #CognitiveAI

🧠 Should students be allowed to use #AI during exams? Colleague Steve Muylle exposes the question behind the question: how do we test what truly matters in an AI age? In Poets&Quants, Steve shares how schools, like ours, can evolve from AI-prohibition to AI-partnership. How to prepare students for a world where human and machine intelligence strengthen one another? 👉 Good read! https://lnkd.in/euiajMkZ #AI #Learning #vlerick

One of the most persistent challenges in creating stateful AI is catastrophic forgetting. How do we teach a model to continuously learn from new data without erasing what it learned before? This is mission-critical for personalization. A recent Google Research paper, "Introducing Nested Learning," proposes a compelling new paradigm that treats this as an architectural problem, not just an optimization one. The core idea is to move beyond a single, uniform learning process. Instead, Nested Learning views a model as a system of interconnected learners, all optimizing at different speeds. Think of it this way:  🔹 Fast learners (outer loops): These parts of the model adapt quickly to new, immediate information. Like a system's cache, they handle the here-and-now.  🔹 Slow learners (inner loops): These deeper parts update deliberately, preserving core knowledge and long-term patterns. This is the persistent storage, the foundation. By creating this "continuum memory system", the model can integrate new information (e.g., a user's new interest) using the fast-updating components, without overwriting the stable, foundational knowledge held in the slower components. It’s a fundamental shift from trying to find one perfect learning rate to building an architecture that has a spectrum of learning rates built into its very structure. This provides a mechanism for more robust and stable continual learning. For anyone working on LLMs that need to evolve over time, this is a paradigm worth watching. It's a step towards models that don't just learn, but remember. Thank you Ali Behrouz and Vahab Mirrokni (et al.) for this fantastic read! https://lnkd.in/dwHzZnsh