Welcome to Verses Over Variables, a newsletter exploring the world of artificial intelligence (AI) and its influence on our society, culture, and our perception of reality.

Back to Basics

Is AI Cheating? AI in Education.

This week, we read about Andrej Karpathy's new venture, Eureka Labs, and it got us thinking about how AI is set to revolutionize education. Karpathy, the ex-head of AI at Tesla and OpenAI co-founder, is building "a new kind of school that is AI-native." But before you start picturing a dystopian future where robots are valedictorians, let's chat about why AI in education isn't the academic apocalypse some fear it to be.

Remember when passing notes was the height of classroom mischief? Well, move over paper airplanes, there's a new troublemaker in town: Artificial Intelligence. But this troublemaker might just be the best thing to happen to education since the invention of the pencil (or at least since calculators were allowed in math class).

The Rise of the Smart Classroom: AI has entered our schools with the subtlety of a cafeteria food fight, but here's the plot twist: instead of turning our students into academic con artists, it's actually making them smarter. Dr. Philippa Hardman, a learning design expert who's become something of an AI whisperer in educational circles, isn't losing sleep over machines taking over the classroom. Instead, she's championing AI as a powerful learning tool. In her work, Hardman emphasizes that AI should be viewed not as a replacement for human educators, but as a means to enhance the learning experience. Hardman argues that AI should be integrated thoughtfully into learning design to improve educational outcomes — she's not talking about letting AI do the heavy lifting, but rather using it to support critical thinking and creativity in education.

Your New Tutor Runs on Electricity: Remember that friend who always seemed to have all the answers? Well, AI is like that friend, but available 24/7 and without needing pizza as payment. Prof. Ethan Mollick, another tech-ed guru from the Wharton School, is particularly excited about AI's potential as a round-the-clock tutor. In his newsletter, "One Useful Thing," Mollick explores how AI can provide personalized and adaptive learning experiences, complementing traditional classroom settings. Mollick envisions a future where AI turns your bedroom into a hi-tech learning lab. Imagine prepping for class with AI at home, then showing up to school to flex your newfound knowledge. It's like having Einstein on speed dial, minus the wild hair.

Meet the New Multimodal Marvel: The latest AI models are like the Swiss Army knives of the digital world – they're multimodal. These high-tech marvels can understand and generate text, interpret images, understand speech, and maybe even help you with your terrible stick figure drawings. These new-age AIs are like having a tutor who's also an art critic, a speech therapist, and maybe even a stand-up comedian all rolled into one. It's not just about answering questions anymore; it's about creating a rich, interactive learning experience that engages all your senses. Learning has never been this sensational – it's a full-blown cognitive carnival, courtesy of our AI ringmasters.

Personalized Learning: Because One Size Fits None: Now, before you start thinking AI will replace our beloved human teachers, pump the brakes. The real magic happens when we combine AI's superpowers with good old-fashioned human intuition and care. Imagine a classroom where AI helps tailor lessons to each student's unique learning style, pace, and interests, while teachers use their irreplaceable human touch to provide emotional support, motivation, and those "aha!" moments that no algorithm can replicate. It's like having a personal learning DJ – AI mixes up the perfect playlist of lessons, while your teacher is the rockstar on stage, bringing it all to life. We’ve already seen some glimpses of the future of AI education with Khan Academy and Eureka Labs.

AI: The Global Education Passport: AI isn't just changing how we learn; it's changing who gets to learn. Imagine a world where a kid in a remote village can access top-notch education in any language, faster than you can say "Google Translate on steroids." AI is turning this sci-fi dream into reality, breaking down language barriers like a linguistic Hulk, and bringing quality education to corners of the globe where textbooks fear to tread. It's not just about translating content; it's about culturally adapting lessons, providing 24/7 tutoring in hundreds of languages, and giving students worldwide a fighting chance at that elusive level playing field. So the next time someone says AI is just for tech bros in Silicon Valley, you can smugly inform them that it's the Robin Hood of education, democratizing learning one algorithm at a time.

Cheating? More Like Leveling Up: But as we open these new doors, we face new challenges. We know what you're thinking: "But won't students just use AI to cheat?" It's a valid concern, but Hardman and Mollick argue that we're asking the wrong question. Hardman suggests a different approach: breaking down assignments into stages and incorporating student reflections. This method can foster a growth mindset and reduce the temptation to use AI as a shortcut. Mollick, on the other hand, acknowledges the challenge but sees it as an opportunity to evolve our educational practices and foster a culture of integrity and engagement among students. Think less "Is AI cheating?" and more "How can we make learning so cool that cheating is lame?" We like to think of AI as a tool — like the web or a calculator. The world of education had to adapt to each new technological advance, but it didn’t collapse.

If You Can't Beat 'Em, Teach 'Em: Let's face it: trying to keep AI out of education is like trying to ban smartphones from a high school hallway. It's already here, quietly revolutionizing how we access and process information. So instead of fighting a losing battle, educators focus on teaching students how to use AI responsibly and effectively. It's less about banning the technology and more about preparing students for a world where AI is as common as email. Conrad Hughes, Director General of the International School of Geneva, puts it bluntly: "We live in a world where architects, designers and engineers use the best technology available to them every day. We must examine whether it is sensible to prevent students from engaging in the same behaviour in their own world."

Welcome to the Future, Where AI Is Your Coworker: Here's a reality check: the working world is embracing AI faster than you can say "update your LinkedIn". According to a recent McKinsey survey, AI adoption in companies has skyrocketed from a steady 50% over the past six years to 72% in 2024. Companies are actively seeking employees who can effectively work alongside AI tools. By integrating AI into education, we're not just preventing cheating—we're preparing the next generation for a workplace where collaboration with AI is the norm, not the exception. That's right, move over bake sales and book fairs, there's a new school partnership in town, and it's powered by algorithms! Schools are shaking hands (metaphorically, of course) with tech companies like Google, Microsoft, and IBM. It's like the ultimate field trip, but instead of visiting the tech company for a day, the tech company is coming to school... every day, turning classrooms into AI literacy boot camps.

AI Literacy: The New Must-Have Skill: Forget learning Python (the coding language, not the snake). There's a new must-have skill, and it's all about AI. We're talking about understanding AI-human teamwork, navigating the ethics minefield, and knowing how to sweet-talk an AI into doing your bidding. According to recent studies, today's innovation-oriented employers are increasingly favoring skills-based hiring and education programs with strong business ROI.

The Final Bell: While 75% of American K-12 teachers are optimistic about the potential for AI in their craft, a whopping 90% feel unsure of where to start, and more than 90% of teachers said they had never had any training in or even advice on how to use generative AI in school. It's like giving a teacher a supercomputer and expecting them to figure it out independently. This gap between enthusiasm and know-how is the educational equivalent of being all dressed up with nowhere to go. If we're going to use this technology (and let's face it, it's not going anywhere), we might as well use it right.

AI in education isn't the end of learning as we know it. It's not a magic wand either. It's more like that exchange student who's shaking things up in all the right ways. With multimodal AI and personalized learning, we're not just upgrading education; we're giving it a total makeover. But remember, at the heart of it all is still the irreplaceable human connection between teachers and students. After all, no AI can replicate that "You've got this!" nod from your favorite teacher.

The New AI Playbook: The bottom line for anyone who cares about the future of education: it’s more Back to the Future than Terminator. i) AI can give your teachers superpowers to help students learn better with customized learning, 24/7. ii) Geography is no longer destiny when it comes to education quality. iii) AI in education is about teaching students how to think, adapt and work alongside smart machines — new skills for a new world.

So the next time someone tells you AI is ruining education, you can confidently say, "Actually, it's just upgrading it." As educators, students, and lifelong learners, it's up to us to shape how AI will be used in education. So, let's roll up our sleeves and start coding... I mean, learning!

We’ll be talking about our favorite tools each week but here is a list of the tools we use most for productivity: ChatGPT 4o (custom GPTs), Midjourney (image creation), Perplexity (for research), Descript (for video, transcripts), Claude (for writing), Adobe (for design), Miro (whiteboarding insights), and Zoom (meeting transcripts, insights, and skip ahead in videos).

Intriguing Stories

The Rise of the Mini Models: AI’s New It Girls: In the world of artificial intelligence, bigger has always been better. Massive language models with billions of parameters have strutted down the runway, dazzling onlookers with their ability to write sonnets, code websites, and pontificate on the meaning of life. But, that's so last season. The hottest new trend in AI: going mini. This week, tech's biggest names unveiled their latest creations: bite-sized AI models that pack a punch without the hefty price tag: GPT-4o mini from OpenAI, Nemo from Mistral, DCLM from Apple and SmolLM. So why the sudden fascination with all things mini? It's not just about looking cute (though that helps). These scaled-down sensations are more efficient, cost-effective, and environmentally friendly than their plus-sized predecessors. They're like the capsule wardrobe of AI – versatile, practical, and with a much smaller footprint.

But what exactly makes these models "mini"? It's about streamlining their complexity – they have fewer parameters, or "neurons," in their digital brains. While the exact details are often kept under wraps, industry insiders suggest they're significantly more compact than the larger models. This digital downsizing translates to reduced computational requirements, allowing these models to process tasks with impressive speed.

Don't be fooled by their size, though. These mini models are surprisingly capable. GPT-4o Mini, for instance, can process an impressive 15 million tokens per minute. That's like speed-reading "War and Peace" while simultaneously drafting a research paper and planning a detailed itinerary – all in the time it takes to finish your morning coffee. The pricing is also attracting attention. At just 15 cents per million input tokens, GPT-4o Mini is making AI capabilities much more accessible. It's bringing high-end AI to a wider audience, much like how ready-to-wear lines bring designer aesthetics to a broader market. This trend also democratizes AI in another critical way: several of these models are open source, allowing for broader access and innovation within the developer community.

The mini model movement shows no signs of slowing down. They're efficient, affordable, and undeniably intriguing. In a world where bigger has always been better, these AI pixies are proving that sometimes, the best things really do come in small packages.

Meta says, Hold My Beer: Meta's largest Llama 3 model, boasting a whopping 405 billion parameters, was leaked Monday morning and is expected to be officially released imminently. This multimodal behemoth dwarfs Meta's smallest 8 billion parameter model, signaling a quantum leap in open-source AI capabilities. Early tests by the developer community suggest it's giving GPT-4o a run for its money, potentially outperforming it on several benchmarks. With its open-source nature, Llama 3 is poised to democratize access to cutting-edge AI technology, though running such a massive model will require serious computational muscle. Llama 3 is more than just a new model; it's a statement. As Meta throws down the gauntlet in the AI race, the landscape of artificial intelligence just got a whole lot more interesting.

Meta Gives EU the Cold Shoulder: Meta announced it won't release its new multimodal AI model in the European Union, citing an "unpredictable" regulatory environment. This decision means European companies won't have access to the open-license model, which can handle video, audio, images, and text. The move follows the EU's recent finalization of compliance deadlines for its new AI Act, giving tech companies until 2026 to meet stringent rules on copyright, transparency, and AI applications. Meta's choice mirrors similar actions by other tech giants, like Apple, who have expressed hesitation about launching certain AI products in the EU due to regulatory concerns. While a text-only version of Meta's Llama 3 model will still be available in the EU, the multimodal version's absence highlights growing tensions between Big Tech and European regulators.

AI Breakout: Researchers and ethical hackers are lobbying for the right to "jailbreak" AI systems—and the U.S. government might just give them the green light. (Jailbreaking, in this context, means manipulating an AI system to bypass its built-in restrictions, often by exploiting vulnerabilities or using clever prompts to make the AI ignore its original instructions.) Unlike smartphone jailbreaking, which typically involves installing unauthorized apps, AI jailbreaking aims to peer into the inner workings of artificial intelligence. These researchers aim to expose biases, reveal training data sources, and uncover potential harms. A proposed exemption to U.S. copyright law would allow experts to circumvent AI companies' terms of service for "good faith research," potentially shedding light on how these increasingly influential systems operate and making them safer and more transparent. However, AI companies aren't standing still. OpenAI, for instance, has already taken steps to fortify its defenses. Their latest model, GPT-4o Mini, introduces a new "instruction hierarchy" technique designed to block one of the most common jailbreaking methods: the "ignore all previous instructions" loophole. This update makes it harder for users to override the AI's original programming, potentially thwarting some research attempts. Whether you're Team Jailbreak or Team Lockdown, one thing's for certain: the future of AI transparency is up for grabs.

Pixel Prints: In a fusion of ancient artistry and cutting-edge technology, Tokyo-based startup Sakana AI has unveiled two groundbreaking AI models that are set to revolutionize the world of ukiyo-e, Japan's beloved traditional art form. Dubbed Evo-Ukiyoe and Evo-Nishikie, these digital artisans can conjure up vibrant woodblock prints at the speed of a bullet train. With a whopping $30 million in funding from tech heavyweights, Sakana AI isn't just painting pretty pictures – they're on a mission to revive ukiyo-e for the Instagram generation. Imagine scrolling through your feed and stumbling upon a meticulously crafted woodblock print of a samurai wielding a selfie stick or a geisha sipping a frappuccino. While the models are still in their infancy, speaking only in Japanese whispers for now, they promise a future where art history and artificial intelligence walk hand in hand down the virtual streets of Edo.

— Lauren Eve Cantor

thanks for reading!

if someone sent this to you or you haven’t done so yet, please sign up below so you never miss an issue.

if you’d like to chat further about opportunities or interest in AI, please feel free to reply.

if you have any feedback or want to engage with any of the topics discussed in Verses Over Variables, please feel free to reply to this email.

we hope to be flowing into your inbox once a week. stay tuned for more!

banner images created with Midjourney.