Sponsored by
15% Off Your First Order of Fresh-Roasted Pods
Most K-Cups sit in a warehouse for months before they reach your brewer — and you can taste it. Angelino's, a third-generation roaster in Los Angeles, does it differently: every pod is roasted, ground, and sealed fresh in-house, then shipped to your door within days of roasting.
Choose from 50+ specialty coffees — single origins, dark roasts, flavored favorites like Jamaican Me Crazy, plus decafs and teas — all compatible with your Keurig® brewer. Mix and match any coffees you like, and watch the savings climb: bulk discounts start at 5+ boxes and reach up to 34% off at 12+.
And here's the best part: new customers get an extra 15% off their first order, on top of bulk savings. It's applied automatically at checkout — no code, no subscription required. Free shipping on orders over $49 or 4+ boxes.
Fresh coffee shouldn't require a coffee shop.
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
380 Trillion Token Study Finds AI Pays Off for Talkers, Not Thinkers
TLDR: A Yale-led study of how people really use AI found that companies tied to AI are beating the stock market, and the jobs gaining the most are built on talking to people, not crunching numbers.
The Story:
Researchers looked at 380 trillion "tokens," the small chunks of text AI reads and writes, from a company called OpenRouter that passes requests to more than 400 AI models. The data ran from January 2024 through April 2026 and covered about 2% of all AI use in the world. They built a weekly measure of how much the planet uses AI, then checked which companies' stock prices rose as AI use grew. Companies with the most AI exposure earned about 0.64% more per week than companies with the least, a gap the team calls the "AI Premium." And it wasn't only tech firms. Stores, consumer brands, and even factory-heavy companies got the bump.
The study also split the win by job type, and the answer surprised the researchers. Work that leans on talking, teaching, and dealing with people scored better. Work built on lab tasks, science, and heavy analysis scored worse. The effect was strongest in the U.S. and Europe and barely showed up in China. One more shift stood out: in 2024, "agentic" AI that runs tasks on its own was a tiny slice of use, but by 2026 it was more than half.
Its Significance:
Here's what it means for your paycheck. The market is betting AI will lift people who are good with people and lean on those whose work is mostly routine analysis. That doesn't mean science jobs are doomed. The researchers point at routine lab work, not the deep, original research AI still can't do. But if your job is mostly communication, coordination, or coaching, the money is drifting your way. AI isn't just a tech story now. Investors are pricing it into stores, factories, and everyday brands, so it may touch your job even if you never open a chatbot. The movement toward agentic AI, tools that act on their own, is the part to watch. Half of all AI use is already there, and the companies closest to those tools are the ones the market rewards most.
QUICK TAKES
The story: Waze is rolling out a "less chatty" setting that trims voice prompts so you can hear your music, plus Gemini voice search that lets you ask things like "find a gas station nearby with the lowest prices." It's also adding a motorcycle mode and letting you report road closures just by talking.
Your takeaway: Your map app is turning into a talking assistant, which is handy, but it runs on Google's AI, and Google itself says not to trust it for safety-critical calls while you drive.
The story: Apple spent more than $10 billion on a self-driving car it canceled in 2024, but the chip work from that project became the Neural Engine that now runs AI on every Mac and iPhone. Bloomberg reports Apple is rushing to its M7 chips, including an M7 Ultra built to hold up to 1.5TB of memory for AI servers.
Your takeaway: A famous flop turned into the backbone of Apple's whole AI push, a good reminder that "failed" projects can quietly pay off years later.
The story: Meta launched "Muse Image," the first product from its new Superintelligence Labs, which let people edit photos by tagging public Instagram accounts, then pulled it within 72 hours after a backlash over using people's images without permission. The actors' union SAG-AFTRA slammed it, saying anything short of a clear opt-in was unacceptable.
Your takeaway: A fancy "superintelligence" label didn't save Meta from a basic privacy misstep, which shows the hard part of AI often isn't the tech, it's thinking about how real people will use it.
TOOLS ON OUR RADAR
🤖 Typebot Freemium: A visual builder that allows you to create high quality conversational chatbots for your website or business by simply dragging and dropping different blocks together like a puzzle.
🎬 OpenToonz Free and Open Source: A highly enjoyable animation program used by professional studios that gives you the creative tools to draw and produce your very own two dimensional cartoons.
🎞️ OpenShot Free and Open Source: A wonderfully simple video editor that provides an easy drag and drop interface so anyone can quickly create beautiful home movies with transitions and titles.
🕹️ ScummVM Free and Open Source: A magical piece of software that breathes new life into classic adventure games allowing you to easily play your favorite retro childhood titles on modern computers.
TRENDING
Feeding AV Software Near-Misses, Not Just Crashes, Makes It 90% Safer — University of Michigan researchers found that training self-driving car software on both crashes and near-misses, instead of crashes alone, lifted its safety performance by 90% in tests at the Mcity track. Near-misses happen about 1,000 times more often than crashes in simulation, so they give the software far more to learn from. It's a step toward proving driverless cars are safe, which matters when about 40,000 people die on U.S. roads each year.
A New Way to Catch AI Models That Make Illegal Images — MIT researchers, working with child-safety group Thorn, built a tool that can tell whether an open-source AI model has been altered to produce illegal content like child sexual abuse material, without ever creating a harmful image. It reads the model's inner wiring and flagged bad models with 100% accuracy in tests. Reports of AI-made CSAM to the National Center for Missing and Exploited Children jumped from 67,000 in 2024 to more than 1.5 million in 2025, so hosting sites could use this to pull dangerous models fast.
Anthropic Keeps Poaching Big Names in AI — Anthropic, the company behind the Claude AI model, has kept up a steady run of high-profile hires this year, including Nobel Prize winner and AlphaFold co-creator John Jumper from Google DeepMind and former OpenAI leader Andrej Karpathy. The pattern shows a company stacking senior talent across research, chips, and science as talk of a possible stock market debut grows.
OpenAI Wants ChatGPT in the Whole Family — OpenAI posted a job for a product manager focused on families, caregivers, and older adults, a sign it wants ChatGPT to be a shared household tool, not just a work app. TechCrunch reports the audience is aging up, with parents' use of ChatGPT nearly doubling in a year. The push comes as OpenAI faces lawsuits from parents over harm to kids and has been adding parental controls and safety features.
DoorDash's AI Assistant Builds Your Cart From a Photo — DoorDash's new "Ask DoorDash" assistant lets you order food and groceries with a typed prompt, a photo of a recipe, or your voice, and it builds the cart for you. Early tests showed carts worth about 35% more than usual and filled five times faster, with nearly half of restaurant orders going to places the customer had never tried. It runs on a mix of models from OpenAI, Anthropic, and Google.
16 Nobel Winners Say AI Could Reshape the Economy Faster Than Anything Before — More than 200 economists and AI researchers, including sixteen Nobel laureates, signed a Stanford-organized statement urging leaders to prepare now for AI's effect on jobs and the economy. They warn that steam, electricity, and computers each gave societies decades to adjust, while AI may give only a few years.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🎤 Name any field. Get the 10 questions an expert would ask that a beginner would never think of.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one API call. Create The Expert Interviewer — a tool that generates the 10 questions an expert would ask in any field. Use localStorage key 'expert_interviewer_v1'.
Aesthetic: near-black (#0d0e14) with a subtle dot-grid overlay, radial vignette. Playfair Display italic 900 for headings, Inter for UI, Courier Prime monospace for labels. Purple/violet (#a09ad0) accent. Question numbers in large italic serif.
Form: field name input, question focus dropdown (6 options: general fluency / hiring / decision-making / spotting fraud / learning from scratch / good vs great), current level dropdown, optional context input.
Call the API instructing it to generate questions a beginner would never think to ask — specific, non-obvious, revealing. Return raw JSON: field_name, expert_edge (2 sentences on what separates expert from knowledgeable amateur), questions array (number, category, question, why_it_matters in 1 sentence), deepest_cut (the single best question), red_flag (the answer that immediately signals a novice or fraud).
Render: a field header with italic Playfair Display field name and focus meta, an expert edge section explaining what separates real experts, the 10 question cards each with large italic number, category label in monospace, the question in large serif, and why-it-matters in small Inter below, a two-column bottom row for deepest cut and red flag. Save sessions to localStorage with field name and date. Make it work in a single HTML file.What this does: Type any field: venture capital, emergency medicine, wine, machine learning, architecture, supply chain. Choose a focus (general fluency, hiring someone, spotting a fraud, learning from scratch) and your current level. The Expert Interviewer generates 10 non-obvious questions that expose real depth, each tagged by category (Mental Models / Trade-offs / Failure Modes / Edge Cases / Judgment) with one sentence on what a strong answer actually reveals. A closing section surfaces the single question that cuts deepest and the most common answer that sounds smart but immediately signals a novice. Every session saves to localStorage so you can build a personal library of expert lenses across fields.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Turn a photo of a recipe or a spoken request into a finished shopping cart with the right items and amounts (DoorDash).
❌ Still Can't: Run fully on its own without a human check. DoorDash still tells users to review the cart before checkout, because the assistant can get things wrong.
✅ AI Can Now: Detect whether an open-source image model was secretly tuned to make illegal content, by reading its internal math, with no harmful image produced (MIT and Thorn).
❌ Still Can't: Reliably avoid making things up. Google warns its in-car Gemini assistant can generate false information, so it shouldn't be your final word on hazards while driving (Waze).
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Pluto (2023) - Anime TV Series
A Netflix animated adaptation of Naoki Urasawa's award-winning manga (itself a reworking of a legendary Osamu Tezuka Astro Boy arc), about a Europol robot detective named Gesicht investigating a series of murders where the victims are the seven most advanced AI robots in the world. Reads like a peak-era Michael Mann film if he directed a robot noir, consistently on lists of the greatest animated series ever made. If you're not into anime, the original manga is well worth reading instead.
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.
Stop Paying for 6 Tools. One AI Does It All.
Most e-commerce sellers juggle 6–8 tools and pay hundreds monthly to keep operations running. StoreClaw replaces the stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks profit 24/7. Connect your store and let AI handle the work — no prompts, no complex setup, no credit card required.