California colleges struggle with online fraud

Really excited to share a recent story in the Wall Street Journal where I talk about the possibilities and challenges of utilizing various digital technologies to enhance school safety. Tech advances including large language models can enhance the accuracy of detection (more true positives AND fewer false positives) likely will increase fidelity of implementation because software isn't constantly pinging staff with alerts that aren't meaningful. This is especially helpful for threats that take place online, including through social media. But at the end of the day, decreasing the threats overall relies on how our communities interact, and how youth treat one another and themselves. A holistic approach is one that balances the need for threat identification tools with strategies that overall reduce the number of threats that exist. Check it out below, and big thanks to Ralph Allora for his reporting and asking me to be a part of this piece! #socialmedia #schoolsafety #tech #safeschools University of Michigan https://lnkd.in/gDKywr_V

🎓 Are Proctored Exams Really Safe — or Just an Illusion of Security? The idea behind proctored exams sounds simple: turn on your camera, share your screen, and let the system make sure no one cheats. Easy, right? Well… not exactly. Modern proctoring tools use AI face tracking, browser lockdowns, and audio detection to spot suspicious behavior. Impressive — but here’s the truth: no system is completely unhackable. Just as technology evolves to catch cheaters, clever minds keep finding creative ways around it. Beyond that, there’s another layer — privacy and fairness. Being watched by a webcam for hours isn’t exactly comfortable, and not every student has a quiet, well-lit space or strong internet connection. So, while proctoring adds security, it can also add stress and inequality. Maybe the real solution isn’t just more monitoring, but better exam design — questions that test critical thinking, not Google skills, and assessments that reward learning over memorizing. So, what do you think — are proctored exams the future of academic honesty, or just a digital illusion of control? 🤔

we, as humans, are good at believing lies. linkedin posts sell you many lies. the most infamous lie on linkedin is that solving dsa pattern-wise can make you good at it. absolutely false. knowing the topic is almost knowing the entire solution. all you're learning is how to apply a technique, and not learning how to think. the only time solving topic-wise is helpful is when you 𝘭𝘦𝘢𝘳𝘯 𝘢 𝘯𝘦𝘸 𝘵𝘰𝘱𝘪𝘤. but after you've done 2 or 3 problems, never solve topic-wise for that topic again. facts above are well known to any decent competitive programmer. to get truly good at problem-solving, you need to learn to figure out 𝘸𝘩𝘢𝘵 𝘵𝘦𝘤𝘩𝘯𝘪𝘲𝘶𝘦𝘴 𝘸𝘰𝘳𝘬 at a problem and 𝘸𝘩𝘢𝘵 𝘵𝘦𝘤𝘩𝘯𝘪𝘲𝘶𝘦𝘴 𝘥𝘰 𝘯𝘰𝘵 𝘸𝘰𝘳𝘬. if i tell you the technique beforehand, you do not 𝘵𝘩𝘪𝘯𝘬. you just 𝘪𝘮𝘱𝘭𝘦𝘮𝘦𝘯𝘵, which is much easier to master and far less valuable.

📱 Are school-issued devices really just for learning? “Throughout the school year, I kept reading about numerous problems with student device use at school, one of which put me over the top,” says Seattle mom, Megan Kaul. Kaul explains that research shows many school-issued devices collect and store enormous amounts of student data. Rather than being designed purely for learning, these tools are often used to feed algorithms that predict children’s futures — and that information can end up in the hands of advertisers, college admissions offices, law enforcement and more. Data breaches are also a serious risk, with one incident in January 2025 exposing 62 million student records and 10 million teacher records. 📖 Read this thought-provoking piece here: http://bit.ly/4gNkM1W

The secret to learning so fast it "feels unfair" is abandoning broken passive methods like rereading and highlighting, which only create familiarity, not memory, as the method itself is broken for 90% of people. The core mechanism for retention is retrieval (the Testing Effect), which mandates actively trying to recall information from memory, recognizing that "No struggle = no growth" is the signal your brain needs to strengthen connections and keep the data. Stop being a Consumer and become a Builder by immediately turning input into output—you remember what you create, so teach the topic, draw a diagram from memory, or write a summary to force reorganization. Maximize focus by trading long sessions for short, hyper-focused sprints (25���30 minutes), using a timer to compress the task and eliminate distractions (Parkinson’s Law), while ruthlessly applying the 80/20 rule to master only the core 20% of concepts that unlock everything else. Ultimately, achieving consistency requires relying on reliable systems—routines that run regardless of how you feel—and adopting the identity: "I’m someone who learns fast". my source is https://lnkd.in/g8qEf36i #LearningHowToLearn #RetrievalPractice #StudyHacks #ProductivityTips #SystemsOverMotivation

Hey guys 👋🏻👋🏻 Why prioritize daily DSA practice in 2025's tech race? Here's how it transforms your skills: Sharpen Your Problem-Solving Edge Daily: In the dynamic world of tech, consistent DSA practice hones your ability to break down complex problems like optimizing algorithms for cybersecurity threat detection turning you into a faster, more intuitive coder who stands out in interviews and on the job. Build Scalable Code Like a Pro: By revisiting data structures (e.g., trees or graphs) and algorithms every day, you'll craft efficient, resource-smart solutions that handle real-world demands, from large-scale data processing in encryption systems to AI integrations, preventing bottlenecks before they arise. Stay Ahead in a Competitive Field: Top firms like Google and Meta prioritize DSA mastery—daily habits via LeetCode or HackerRank prep you for coding challenges, boosting your hireability and adaptability to innovations like quantum-resistant computing. Turn Theory into Superpower Habits: Just a few hours a day reinforces logical thinking, making DSA second nature for everyday decisions, whether debugging cryptography flaws or scaling secure apps, ultimately elevating your career trajectory with confidence and creativity. Which tip will you implement today? #DSA #TechSkills #CodingDaily 

LLMs struggle with “I don’t know” for the same reason smart people do. OpenAI’s new white paper shows why: hallucinating is statistically rational. On the current SAT, there’s no penalty for guessing—so you should always guess. On the old SAT, wrong answers cost you points—so restraint was smarter. Today’s LLMs are basically stuck in “new SAT mode.” Top students and young associates often fall into the same trap. They’ve been rewarded for being right so often that they never build the muscle of calibrating confidence. Overconfidence works—until it doesn’t. That’s why we spend so much time training new investors to get comfortable saying, “I don’t know.” It builds trust. LLMs can’t do that yet. But you can bridge the gap with prompting: add “If you don’t know, don’t guess” and watch hallucinations drop. Attaching a screenshot of my specific prompt. Just like the SAT eventually changed its rules in 2016, LLMs will one day learn to penalize overconfidence too. Until then, prompt design is the best workaround. If you’ve found other prompt hacks that reduce hallucinations, I’d love to hear them.

Recently I found a surprisingly effective way to learn almost any topic in a reasonable amount of time. Here’s my approach: 1️⃣ Start with LLM. I ask it to tell me everything I need to know to become an expert in a new field (with "deep research mode") and pick a few topics that sound interesting, explain them in simple language with detailed examples, preferably by connecting them to something I already understand. That helps me build a big-picture map first. 2️⃣ Dive deep and go official. Once I have a mental framework, I read the official docs, tutorials, or papers and follow along step-by-step to see how things are formally done. 3️⃣ Build something that could be be verified. I immediately start a small project or problem related to what I just learned. That’s where the real understanding forms, when I get stuck and have to figure things out. What’s interesting is that this process is similar to how most of current LLMs are trained: Pre-training with web-scale data → Supervised Fine-Tuning with high-quality data → Reinforcement Learning from Verifiable Feedback (RLVF).

Just read Education Week’s piece, “Schools Face an Uphill Battle in Protecting Student Data in the Age of AI” (March 2025) — and it really hits home. As districts move more student records online, keeping that data safe has never been more important. That’s why at YellowFolder, we focus on making digital record management both secure and simple — with encrypted storage, permission-based access, and automated retention. Going digital should make life easier and keep data protected. 💛 #K12 #EdTech #DataSecurity #DigitalTransformation #YellowFolder https://lnkd.in/gse2viDJ

In-Context Learning Isn’t Memorization When an LLM performs few-shot tasks from examples in the prompt, it’s not recalling stored weights. It’s dynamically constructing a temporary function within the attention pattern—an ephemeral program. This transient computation lives only for the duration of the context window. That’s why scaling context length increases apparent “reasoning” capability: it extends working memory, not intelligence.