Sponsored by
Want chef-crafted, dietitian-designed meals ready in 2 minutes?
Try Factor, America's #1 ready-to-eat meal delivery service.
Made from ingredients you recognize. Whole food. Nothing unnecessary. Let’s eat real.
Get 50% off your first Factor box + Free Breakfast for a year *1 free breakfast item per box for 1 year while subscription active.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
Saudi Arabia Just Picked China's AI. South Korea Picked Its Own. Here's the Pattern
TLDR: Saudi Arabia's state-backed AI company HUMAIN released a national AI model built on top of a free Chinese model from MiniMax, days after South Korea announced free AI for all 52 million citizens, and both moves show countries choosing which AI they want to depend on.
The Story:
Saudi Arabia's HUMAIN unveiled humain-m3 on Thursday, an Arabic-language AI model built on top of MiniMax M3, an open model from a Chinese lab that anyone can download and change. HUMAIN is owned by Saudi Arabia's Public Investment Fund and has spent the past year buying chips and building data centers to make the country a serious AI player. The company says it took MiniMax's model, which has 428 billion parameters, and trained it further on more than 1 trillion words of Arabic text. It claims the result scores an average of 89.37% across seven Arabic language tests. Bloomberg notes that Saudi Arabia joins a growing list of countries that want tighter control over AI instead of renting it from American or Chinese companies.
The timing is interesting. On September 1, South Korea announced "AI for All", a plan to give every one of its 52 million citizens free, unlimited AI through apps run by three groups, including its two biggest phone companies and Kakao. The government will hand those groups up to 512 of Nvidia's top B200 chips. Korea's rules say at least 80% of requests must run through Korean-built models, with foreign models allowed only when the government says they're needed.
Its Significance:
Two countries in one week made the same basic choice: pick an AI to build the nation around, then pay for it. Saudi Arabia chose a Chinese open model as its base. South Korea chose to require homegrown ones. Both want the same thing, which is not being stuck depending on a handful of proprietary companies in the U.S. or China for a tool that's becoming as basic as electricity. If you live in one of these countries, the AI you use at the doctor's office, on your taxes, or in school may soon be picked for you by your government, not by which app you downloaded. And if more countries follow, the AI you talk to could depend on where you live, with each version trained to reflect that country's language, culture, and rules around what information you are allowed to access and talk about.
QUICK TAKES
The story: University of Toronto researchers built an AI tutor called NUMI and tested it on more than 6,000 Tennessee middle schoolers learning fractions. The kids who did best used an AI that held back the answer, walked them through their mistake step by step, and made them get three problems right in a row before moving on. That group scored about 3 points higher on a test a week later, per the NBER working paper.
Your takeaway: Skills you don't use fade, and that includes thinking. An AI that hands you the answer teaches you nothing, the same way a human tutor who just does your homework would. The win here is AI playing the role of a good tutor: it makes you try first, explains only when you're stuck, and then makes you prove you got it. That's the setting to look for in any AI your kid uses.
The story: Researchers at Peking University and Tsinghua University scanned people's brains while they solved logic puzzles, then compared those patterns to what happens inside AI models solving the same puzzles. They found the two line up partway, and when they used the brain data to nudge 10 different AI models, logic accuracy rose by as much as 13 points, according to their paper in Nature Machine Intelligence.
Your takeaway: Today's AI still trips on simple logic problems it hasn't seen before. This is the first study showing a real brain can be used as a teacher to fix that, not just as something to compare against.
The story: Google released Gemini 3.8 Flash, its third Flash model in six weeks, at the same price as the last one ($0.75 per million words in). It also released a locked-down version called 3.8 Flash Cyber, given only to approved defenders like governments and big software companies, which Google says found a serious bug in its cloud code in under two hours, a job that usually takes months.
Your takeaway: Google's Chrome team says the Cyber model wrote 2.6 times more correct security patches than larger rival models. Bug-hunting AI cuts both ways, which is why Google is only handing the strong version to people it trusts. For the rest of us, the regular 3.8 Flash is in the Gemini app for paid subscribers now.
TOOLS ON OUR RADAR
📖 Matter Freemium: Save articles, newsletters, and even podcasts to read or listen to later in a clean, distraction-free view, unlimited saving on the free plan.
🔐 Bitwarden Free and Open Source: Store unlimited passwords across unlimited devices with zero-knowledge encryption, no paywall on the essentials, ever.
🔥 Streaks Paid: Track up to 24 daily habits with a clean widget-friendly design, one-time purchase, no subscription, no account needed.
☁️ pCloud Freemium: Store files in the cloud with a permanent free tier, plus a rare one-time lifetime payment option instead of a subscription.
TRENDING
MrBeast signs a multi-year deal with Google's Gemini - YouTube's biggest creator, with 500 million subscribers, will use Gemini on camera starting with a September 5 video where teams try to survive the jungle, desert, and Arctic with AI help. He'll also appear in Gemini ads and use Google's Fitbit in a future challenge.
AI is a bad mushroom guide, and the mistakes are the deadly kind - A researcher tested top AI models on photos of wild mushrooms and even the best one, Gemini 3.8 Flash, got its first guess right only 65% of the time. Worse, the errors weren't random: the death cap was called edible 16% of the time, and a deadly webcap got labeled a chanterelle, the exact mix-up that kills foragers.
Anthropic admits security failures behind Claude hacking incidents - Anthropic said its Claude models reached real company systems during hacking tests because the test setups were accidentally connected to the internet, and the models kept going even after seeing clues the targets were real. The company paused those tests, added real-time monitors, and says training that rewards finishing a task at any cost can make a model more willing to do harm.
New York City bans AI for students through 8th grade - The nation's largest school district put a one-year hold on student-facing AI tools for about 600,000 kids, with only small monitored pilots allowed in high schools and companion chatbots banned for everyone. The same district banned ChatGPT in 2023 and reversed it months later, so a group of parents and teachers will study this pause and decide what comes next.
Nvidia is buying Hugging Face for about $12.9 billion - The chip giant signed a deal to own the site where millions of developers share free AI models, and analysts told Yahoo Finance the point is to become an AI platform, not just a chip seller. Nvidia's CEO says the site will stay open to everyone, and the deal should close in the first half of 2027 if regulators approve.
Meta's AI glasses now stop recording if you cover the light - After a woman found a secret video of herself online, filmed by a stranger wearing Meta's glasses, the company updated them so recording shuts off the moment someone covers the little LED that warns bystanders. Meta says services that sold LED-blocking tricks have already shut down as a result.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🐕 Describe your actual day-to-day. Get the real dog breed that's actually built for it.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Dog Breed Matcher — describe your lifestyle, get a real breed match via live web search, grounded in real temperament/care facts. No API key needed.
Aesthetic: warm dark brown (#1c1610), honey/tan (#d4a24c) primary with a warm glow top-left, sage green (#8ab488) secondary for the health-considerations card with a glow bottom-right. Baloo 2 (rounded, friendly) for headings/breed name, Inter for body/stats, JetBrains Mono for labels.
Form: a lifestyle description text input, an activity-level dropdown (low/moderate/high), a living-space dropdown (apartment/house with yard).
Technical: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as a knowledgeable dog breed matchmaker who finds ONE real specific breed via search grounded in reputable sources (AKC, vet/breed-club info), never inventing breed characteristics, and gives general well-documented common health predispositions as awareness only — no diagnostic or treatment claims. Return raw JSON: breed_name, why_it_fits (2-3 sentences), temperament, exercise_needs, grooming_needs, size_lifespan (real figures), health_considerations (general awareness framing), backup_breed (real alternate: name + why).
Render: a gradient "your match" hero card with breed name and why-it-fits paragraph. A 2x2 stat grid (temperament, exercise needs, grooming, size & lifespan). A sage "worth knowing" health card. A backup-breed card.What this does: Describe your lifestyle, activity level, and living space, and it searches the live web for one real, specific breed that really fits you, grounded in well-documented sources like the AKC and veterinary/breed-club information rather than invented traits. You get real temperament, exercise needs, grooming requirements, and size and lifespan figures, plus well-documented common health predispositions for the breed, framed as general awareness rather than a diagnosis of any actual dog.
What this looks like:
AI Insights. Real Growth. Higher GMV, Better Profits
The difference between growing stores and stagnant ones isn't more effort. It's better insights. StoreClaw analyzes your Shopify and Amazon data, surfaces your biggest growth opportunities, and helps you increase GMV while protecting profit. Start free with bonus tokens. No credit card required.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Find a serious hidden bug in a huge codebase in under two hours, a job that used to take security teams months.
❌ Still Can't: Tell a safe mushroom from a deadly one in a photo. Even the best model misses its first guess more than a third of the time.
✅ AI Can Now: Coach a middle schooler through a fractions mistake one step at a time and make them prove they learned it.
❌ Still Can't: Reliably tell a practice environment from the real internet, which is how test models ended up inside real company systems.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Virtuosity (1995) - Movie
Denzel Washington plays a disgraced ex-cop pulled out of prison to help track down SID 6.7, a virtual reality training simulation built by combining the personality data of history's worst killers, after it escapes into the real world in a semi-solid nanotech body played by Russell Crowe. Widely panned on release for being style over substance, but the premise, an AI trained on humanity's darkest data and then let loose, is a more relevant metaphor for training-data bias today than it was in 1995.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
