Interactive Learning Methods

I received a PhD in Machine Learning from MIT in 2022. 6 months back, I started a project to teach “Hands on Large Language Models” to students, researchers and industry professionals. The result is a mega-project with 30 videos covering everything about using LLMs practically. I have uploaded all videos on Youtube. Lecture 1: Hands on Large Language Models: Series Introduction  https://lnkd.in/gdBbBm62   Lecture 2: The LLM Evolutionary Tree https://lnkd.in/gUcJbqEh Lecture 3: Running Microsoft Phi-3 LLM using Hugging Face | Hands-on deployment https://lnkd.in/ghcp2JbW Lecture 4: BERT + Finetuning model for Movie Review Sentiment Analysis https://lnkd.in/gR_DhR4h Lecture 5: Implementing Flan-T5 Generative model for movie review classification https://lnkd.in/g2BnHdR8 Lecture 6: Using ChatGPT API for Movie Review Classification | Hands-on Project https://lnkd.in/gjJgdWNa Lecture 7: Text Clustering using Sentence Transformers | Hands on Project on ArXiV research paper abstracts https://lnkd.in/gU-yJMah Lecture 8: Topic Modeling using BERTopic | Hands-on projects using ArXiV research papers dataset https://lnkd.in/gti7-WiU Lecture 9: Large Language Models for Text Clustering and Topic Modeling https://lnkd.in/gYiH5PFP Lecture 10: Introduction to Prompt Engineering. https://lnkd.in/gd4g2KxE Lecture 11: Advanced Prompt Engineering | In-context learning | Chain of thought | Tree of thought https://lnkd.in/gmehuvmZ Lecture 12: LLM Guardrails: How to control LLM output https://lnkd.in/g9jTafzZ Lecture 13: Langchain and Agents Introduction https://lnkd.in/gP_GUT6g Lecture 14: What is LLM Quantization? https://lnkd.in/g8pcHVmf Lecture 15: Coding chains with Langchain | Hands-on demonstration https://lnkd.in/gDJ5xASG Lecture 16: How to give memory to Large Language Models? https://lnkd.in/gQBZxtGb Lecture 17: Code your first LLM agent using LangChain https://lnkd.in/gufpKC2Z Lecture 18: Semantic search and RAG https://lnkd.in/gigKxtF2 Lecture 19: Coding a LLM Dense Retrieval System https://lnkd.in/gp9wSVzD Lecture 20: Chunking Strategies for Large Language Models (LLMs) https://lnkd.in/gcCYAKje I have spent a lot of time and effort in making these lectures. I show everything on a whiteboard and then show it through Python code. Want to learn about all above aspects live with me as the course instructor? Register for the Live Generative AI Bootcamp here: https://lnkd.in/gSJuV9sK The course instructors are MIT PhDs and we have live classes every weekend! DM me if you have any queries or questions.

A neuroscientist in Norway spent 20 years testing what your grandmother already knew. Audrey van der Meer, a professor at NTNU in Trondheim, put 256-electrode EEG caps on students and gave them a simple task: Write words by hand. Then type the same words on a keyboard. The brain did not respond the same way. Handwriting set off a coordinated burst across the brain. Memory, touch, movement, and visual areas worked together. Theta and alpha rhythms rose across parietal and central regions, the kind of activity linked to learning. Then the students typed. Much of that network went quiet. Typing relies on repeated, identical keystrokes. The brain has less to map, less to track, less to remember. Other research points the same way: ↳ Princeton found that students who took notes by hand did better on tests of conceptual understanding ↳ Laptop users captured more words, but processed less ↳ Adults recalled calendar events about 25% faster when they wrote them by hand ↳ Print and cursive both worked. The varied hand movements mattered most We gave students keyboards because they were faster. But faster input can mean thinner thinking. The brain does not treat every movement the same. Forming a letter by hand is a small spatial act. Pressing a key is not. Some U.S. states are bringing handwriting back into schools. Van der Meer's team is also urging educators to protect handwriting in early education before tablets take over. One answer to modern learning may not be another app. It may be a pen. What do you still prefer to do by hand? Follow me, Dr. Martha Boeckenfeld, for ideas on thriving as AI rises and leaders stay human. Source: van der Meer, A. L. H., & van der Weel, F. R. R. (2024). Handwriting but not typewriting leads to widespread brain connectivity. Frontiers in Psychology, 14, 1219945. milamarksonoffice

In an age of screens and swipes, it’s easy to forget the power of the pen. But research shows that writing by hand is one of the most effective tools we have for thinking clearly, remembering deeply, and learning faster. Why? Because handwriting engages the brain differently than typing. It activates multiple senses at once, visual, auditory, and kinesthetic, creating stronger memory pathways and deeper comprehension. It’s not just about penmanship. It’s about cognition. One study found that college students who hand wrote their notes retained more than those who typed. Another found that children who formed letters by hand learned them more effectively, and became better readers and writers over time. And for students with dyslexia or dysgraphia, cursive writing can help ease letter reversals, build rhythm, and strengthen fluency. The takeaway: handwriting is far from outdated. It’s a mental workout, especially for young minds still learning to read, write, and make sense of the world. Let’s not write it off. 

So...can Claude Design replace Storyline’s view mode and try mode for software simulations? That was the question I wanted to answer this week. If you’ve ever built software training in Articulate Storyline, you know software simulations can be incredibly effective learning tools. You also know that Storyline’s view mode and try mode can be frustrating to create and maintain, especially once you move beyond simple click interactions. Text entry fields can be particularly painful. Small interface changes often require re-recording screens, rebuilding hotspots, adjusting feedback layers, or recreating parts of the simulation altogether. And when software interfaces are changing right up until launch, maintaining those simulations can become a project all by itself. What made this experiment particularly interesting is that it actually started with Anthropic’s new Fable 5 model. Before it was pulled offline, I had a chance to test it by providing a series of screenshots from a fictional CRM platform along with a simple description of the workflow I wanted learners to complete. With a single prompt, it generated both a guided software demonstration and an interactive simulation where learners could practice the workflow themselves. It included animated cursor movements, on-screen callouts, feedback messages, hints, text-entry validation, and even offered to package the experience as a SCORM course. That immediately raised another question: If Fable 5 could do this, could I recreate something similar using Claude Design? So I opened Claude Design, uploaded the same screenshots, described the workflow, and started experimenting. After a few rounds of refinement, I had a working software simulation that included a guided demo mode, an interactive practice mode, feedback for incorrect clicks, hints after multiple attempts, synced audio narration, and a standalone HTML file that could be hosted on the web. What surprised me most wasn’t necessarily that it worked...it was how quickly I was able to iterate. Instead of recording screens and building interactions slide by slide, the workflow felt much closer to editing and refining. If a screenshot changed, I could update the project rather than rebuilding significant portions of it. So, do I think this replaces Storyline today? Not entirely. There are still important questions around accessibility, SCORM, LMS tracking, governance, collaboration workflows, and long-term maintenance. But each time I run one of these experiments, I find myself less focused on whether AI perfectly replicates existing authoring tools and more focused on how these workflows might fundamentally change the way we create learning experiences in the future. 🔗 Watch the full experiment here: https://lnkd.in/g9SpjHaj #InstructionalDesign #eLearning #LearningAndDevelopment #ClaudeDesign #AI #ArtificialIntelligence #eLearningDevelopment #ArticulateStoryline #VibeCoding #Fable5

Today, our favorite chatbot Claude has gotten an upgrade. Anthropic is going all in to create a GenAI Chatbot that is for everyone from developers to teachers with their artifacts feature. Anthropic just widely released artifacts for all Claude users across their free and premium plans, as well as in their iOS and Android apps. Artifacts is a game-changer for educators, creating a dedicated window alongside the chat where Claude takes your written prompts and turns them into interactive games, presentations, websites, etc. Claude also has made it easy to publish the artifacts, so students or colleagues can interact with the activities. Mandy DePriest, on our team at AI for Education, recently demonstrated how the artifacts feature works with a quick walkthrough of a common classroom use case: creating an interactive, digital vocabulary quiz in seconds with no code and only a few simple prompts. In our training on Monday with a district in NJ, I modeled creating an interactive game out of the popular Egg Drop Challenge STEM project, adding a Batman themed twist to help students in their mission – trust me when I tell you teachers were super excited to start building their own interactive elements. If you want to try artifacts out, here are some example prompts to get you started: -Create a webpage for a high school English class based on this uploaded syllabus (we love the upload feature on Claude) -Generate an interactive math game to help 4th-grade students master comparing fractions. Use faction bars when providing students explanations for the questions -Design an interactive game to help students' learn Newton's First Law of Motion. Include an interactive element and directions for the game -Create an interactive vocabulary quiz based on keywords in the attached file We're excited about the potential of the artifacts feature to help teachers create fun and interactive ways to engage students in the classroom. Add it to the list of the many reasons to try out Claude. Have you used Claude and the artifacts feature yet? Share your best results! Links in the comments to my game and our Claude video walk through. #aiforeducation #GenAI #Claude #teachingwithAI

Patterns.dev - our free project on React + JS design patterns has been updated! Now used by 1M developers since we launched! Over the last while we have gone through the React guides on patterns.dev and brought them in line with how teams are building apps in 2025, with a strong focus on React 18 plus, React 19 and beyond, and the Next.js App Router. Happy to share 1 million developers have used the site since we launched. Here is what changed: Modern React patterns by default We have updated the core React overview and patterns to favor function components and Hooks over classes. The Container and Presentational, HOC, and Render Props patterns are now framed around custom Hooks and composition rather than wrappers that deepen your tree or create callback pyramids. The guidance now reflects how people actually share logic today: Hooks first, HOCs and render props only where they truly make sense. Better Hooks guidance for real apps The Hooks section has been refreshed to align with the latest React docs. There is more emphasis on avoiding unnecessary effects, doing more work in render and event handlers, and preparing for newer APIs like useEvent and the React optimizing compiler so you can worry less about manual memoization. Rendering patterns for today’s React and Next.js The rendering chapters have been updated across the board. Client side rendering, SSR, static generation, ISR, progressive and selective hydration, streaming SSR, and React Server Components are all revisited through the lens of React 18 and the Next.js App Router. The content now leans into streaming, Suspense, server components, incremental static regeneration, and partial prerendering so you can choose the right mix of CSR, SSR, and SSG for your app rather than defaulting to pure CSR. Next.js and full stack patterns The Next.js overview now centers on the App Router, server components by default, async data fetching in server components, generateStaticParams, and modern ISR and cache invalidation patterns. We also touch on server actions so that many mutations can stay on the server instead of living in extra client side code. We have a new write-up on AI-centric user interfaces and how to leverage React and the AI SDK to build them leveraging the modern patterns we talk about on the site. Patterns that aged well stay, with small refinements Some patterns, like compound components powered by context, are still very much recommended. Those sections are now tuned with minor improvements around performance, ergonomics, and compatibility with newer React features. If you are teaching, reviewing, or modernizing a React or Next.js codebase, I hope this refresh makes patterns.dev a useful reference for the next wave of best practices. #softwareengineering #programming #javascript

HOW we work with AI matters. Emerging modes of interaction are reshaping roles. Most people are stuck at method 1 or 2. Here are 4 key AI interaction types—and when to use each: 1️⃣ AI as a Microtasker One-shot problem solver. Ideal for quick, contained tasks: rewriting a sentence, generating a one-off image, answering a data question, or fixing a bit of code. High precision, low overhead. 2️⃣ AI as a Copilot Persistent, live support for extended tasks. It stays with you in pairing mode—watching your screen, listening, coding, brainstorming. A back-and-forth partner for creative or technical work in real time. Human in the loop, always. 3️⃣ AI as a Delegate Assign it a goal and let it work autonomously, for minutes or days. Great for complex, long-form tasks like research—no human in the loop. It self-directs, self-checks, and reports back after/while completing tasks. Think: Manus AI, autonomous agents. 4️⃣ AI as a Teammate A presence across your team or org. It joins meetings, takes notes, surfaces insights, runs simulations, offers opinions. Can even be in a manager role. Not just assisting YOU but enhancing the collective. An ambient, participatory AI system. And roles 3 and 4 mean the AI can work in a completely different way than our human systems. Knowing which role to use—and when—is the new AI literacy.

If you’re an aspiring ML or AI engineer, here are 5 projects you can build right now to get your hands dirty with the latest tools, and actually understand how things work end-to-end 👇🏽 🛠️ 1. Build an ML Pipeline from Scratch Project: Predict which students are at risk of dropping out → Pull attendance + grades → clean in Pandas/Polars → Train with LightGBM → track with MLflow → Deploy via FastAPI + Docker → push to AWS Lambda → Set up CI/CD with GitHub Actions 💡 This is a full ML lifecycle in one project. 💬 2. Build a RAG Chatbot Project: A chatbot that answers questions from course notes → Chunk + embed using LlamaIndex / Cohere / VoyageAI → Store in Weaviate or FAISS → Use an open-source LLM like Mistral or Qwen → UI in Gradio or Streamlit 💡 Great hands-on intro to how GenAI apps actually work. 🧠 3. Fine-Tune an LLM for a Domain Project: A medical or finance assistant → Collect PubMed papers or SEC filings → Fine-tune with QLoRA/DPO → Start with Mistral or DeepSeek Coder → Evaluate on factuality + accuracy → Deploy with Modal or Fireworks AI 💡 Helps you go deeper into foundation model adaptation. 🚨 4. Monitor for Model Drift Project: A fraud detection system that stays effective post-launch → Serve with BentoML → Track drift using Evidently AI or WhyLabs → Use Airflow/Prefect for retraining → Visualize everything in Grafana 💡 Most ML projects die post-deployment, this one shows you can maintain. 🍽️ 5. Build a Multimodal AI App Project: Snap a photo of food → get nutrition + recipes → Use Florence-2 or CLIP for vision embeddings → Add Qwen-VL or LLaVA for reasoning → Build the UI with Streamlit 💡 You’ll learn how to mix modalities and push boundaries. None of these are “easy,” but that’s the point. They’ll teach you the actual workflow, data prep → modeling → deployment → monitoring. And they’ll stand out when someone looks at your GitHub or portfolio. If you end up building one of these (or already have), drop a link in the comments, I’d love to see it 🙌🏽 〰️〰️〰️ Follow me (Aishwarya Srinivasan) for more AI insight and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

I interviewed a dentist who believes that if we trained pilots the way we train dentists, they would all be dead. Lincoln Harris shared this provocative insight on the show today. His reasoning is simple: You cannot land a plane safely after just reading a book or watching one demonstration. You need thousands of repetitions in a simulator. But in dentistry, and often in medicine, we graduate professionals who have practiced limited repetitions on plastic teeth in artificial environments, then send them out to learn on real patients. Even continuing education often happens in hotel conference rooms, far removed from the reality of a clinic. Dr. Harris is changing this with cloud-based simulation training. It is essentially a flight simulator for dental surgery. Instead of flying to a teaching institute, dentists receive specialized mannequins to use in their own operatories. They practice complex procedures using their own chairs, their own lights, and their own instruments. They then upload high-resolution photos of their work to the cloud, where experts from anywhere in the world provide detailed feedback. The results are undeniable. Dr. Harris told me that while traditional courses have an implementation rate of about 15 to 18 percent, his method sees dentists implementing new procedures at four times that rate. It costs a quarter of the price and removes the need for travel. This isn't just about better fillings. It is about a fundamental shift in how we train surgeons. As Lincoln put it, everyone wants their pilot to have thousands of hours of experience, not just to have read thousands of papers. We should demand the same from our healthcare providers. 🎙️ Listen to "Why modern dentists must train like pilots" on The Podcast by KevinMD. (Link in the comments ⬇️) #KevinMD #Dentistry #MedicalEducation #PatientSafety #HealthTech

Most corporate training follows this pattern: - 3 days of training. - Hundreds of slides. - Polite feedback forms. And almost zero change in behaviour. I once looked at a programme that had: • 16 hours of lectures • 6 hours of discussion • A few “reflection activities” And when people went back to work on Monday? Nothing changed. -Not because the facilitator was bad. -Not because the participants were lazy. -Because the learning design was broken. Here is the uncomfortable truth about training: -People do not learn from listening. -People learn from doing. So I started using a very simple rule when designing workshops. The 3–30–300 Rule. 3 minutes → Explain the business problem 30 minutes → Teach the key skills 300 minutes → Practice in real work That is it. Most programmes invert this. They spend 300 minutes explaining concepts and 3 minutes asking people to apply them. Then everyone wonders why nothing sticks. But the moment you flip the ratio, something powerful happens. -People stop being passive participants. -They start becoming active problem solvers. They practice. They experiment. They make mistakes. They improve. And suddenly learning starts showing up where it matters: At work. So the real question every L&D professional should ask is this: If this training disappears tomorrow, will performance actually drop? If the answer is no, the programme was probably just information. Not learning. I turned this thinking into a simple visual framework. Take a look at the infographic below. And I am curious: How much of your training time is spent on input versus application? Let me know in the comments. ___ Save this for later (three dots, top right). Share with friends → ♻️ Repost. ----- If you need corporate learning support, let me know! ----- For more such ideas/content, follow me: Zubin Rashid ----- #LearningAndDevelopment #TalentDevelopment #CapabilityBuilding #PerformanceImprovement #StrategicLnD #Upskilling #Reskilling #BusinessAlignment #WorkforceTransformation #ContinuousDevelopment #LeadershipGrowth #EmployeeGrowth #LearningStrategy #SkillsDevelopment #HRStrategy #OrganizationalAgility