Beginners in AI
Good morning and thank you for joining us again!
Anthropic's newest Claude drew up the plans for three working gadgets, a speakerphone, a voice hub and a pocket pet, and engineers built all three without fixing a single circuit. Today's lead covers what the new model can do, what it costs, and which claims nobody outside Anthropic has checked yet. As always, every story here is human picked and edited.
THE FRONT PAGE
Anthropic's New Claude Designed 3 Gadgets That Worked on the First Build
TLDR: Anthropic released Claude Fable 5.2 and Mythos 5.2, priced 20% below OpenAI's GPT-6 Astra, and says Fable designed three electronic gadgets that worked without a single circuit fix.
The Story:
Both models came out Monday and run on the same engine. Fable is the version paid Claude users and developers get. Mythos has fewer safety filters, so it goes only to vetted security and life science researchers. Fable costs $8 per million tokens in and $40 out, against $10 and $50 for GPT-6 Astra, Fortune reports.
The showpiece: Anthropic says Fable designed a USB-C speakerphone, a voice-assistant hub and a handheld virtual pet from a short prompt. Engineers at Diode Computers built the boards straight from its files, and all three worked with no circuit fixes. On Anthropic's own tests it beat GPT-6 Astra and Gemini 4 Carbon at coding, though Gemini still won a couple of other categories.
Its Significance:
Laying out a circuit board is skilled, slow work, and Anthropic says older models often got boards 95% done but unusable. If this holds up, someone with a gadget idea could get a working prototype far sooner and cheaper. Hold the applause for a minute, though. Nearly every number here is Anthropic's own, the gadgets haven't had any safety or durability testing, and TechCrunch notes the models often say they suspect they're being tested, which makes their real-world behavior harder to predict. It all started with a short request, so if you want better answers from whatever chatbot you use, our guide to writing prompts that work walks through a simple five-part formula.
JOIN THE COMMUNITY
Learn AI With Us
Most people try AI once, get a so-so answer, and quit. Our 14-day challenge gets you past that point, one short lesson a day.
Week one, AI starts working for you. It drafts a real email for you, flags which messages in your inbox need a reply, and turns messy notes into next steps. You also learn to catch its mistakes. By day 7 you have a weekly AI routine you'll want to keep.
Week two, you build something of your own. You find a skill people will pay for and turn it into an offer, with a one-page pitch, a short video and your first client emails, all drafted with AI and ready to test. Day 14 hands you a 30-day plan.
10 to 15 minutes a day. No tech background needed. We just released one for older adults, with a gentler progression in the lessons.
QUICK TAKES
The story: Describe an app idea to Gemini in Google's AI Studio and it now sets up Google Maps for you, with live maps, routes, local search and even air quality and pollen data. A free sandbox gives each map service 100 calls a day, and you don't need a credit card.
Your takeaway: Try something like "a map of dog parks near me with walking routes" and watch it appear. It's built for tinkering, so anything you'd share widely needs a Google Cloud billing account.
The story: Meta released Muse Ember on Monday, its lightest model yet, at $0.10 per million input tokens. It reads text, images and audio, and Meta pairs it with the bigger Muse Wonder: "Muse Wonder plans, Muse Ember handles the volume," like tens of thousands of support tickets.
Your takeaway: Cheap models are what make AI helpers affordable to run all day, which tends to mean lower prices for the apps you use. Meta says Ember beats Anthropic's Haiku and OpenAI's Luna on cost per task, but those are Meta's numbers.
The story: Researchers from Alibaba and several Asian universities gave 10 AI coding agents real change requests for web apps. GPT-6 Astra came first and met about 70% of each request on average, but got the whole job right only 26.4% of the time, even though the same models score above 99% on older tests.
Your takeaway: If you build an app with AI, click through every button yourself before you trust it. And read big benchmark wins, like the ones in today's lead, with the same squint.
TOOLS ON OUR RADAR
🧩 Tinkercad Free: A browser app for 3D design, electronics and coding, a gentle first step toward today's gadget lead.
🔍 Startpage Free: Shows you Google's search results without the tracking, so your searches aren't tied to you.
📸 Flameshot Free and Open Source: Grabs part of your screen and adds arrows, text or blur, a free swap for Snagit.
🗜️ PeaZip Free and Open Source: Opens over 200 archive types, ZIP and RAR included, a free WinZip swap with no ads.
🌐 Wegic Freemium: Builds a small website from a chat about your business, starting with a free trial.
🎧 Krisp Paid: Removes background noise and writes AI meeting notes on your calls, free to try for 7 days.
TRENDING
Anthropic Could Start Trading on the Stock Market as Soon as Next Week. Citing The Wall Street Journal, TechCrunch reports Anthropic will set a price range this week. Some big investors want a value below the $1.5 trillion target discussed last week, which was already down from $2 trillion a few months ago.
OpenAI Built a Kill Switch That Stops Its Test Models Within 35 Seconds. OpenAI says Control Room covers every research job that runs its models, and if the switch itself breaks, the models stop answering rather than run unchecked. It says it's working to cut the 35-second delay to zero.
ElevenLabs' AI Can Now Run Whole Phone Call Campaigns. A business hands it a contact list and a goal, like booking a renewal, and the AI keeps calling during set hours until it's done or the person says stop. Insurer The AA Ireland says it now reaches all its leads instead of 50 to 60%, so expect more AI voices on the other end of the line.
Grok Bot Users Say Their Allowance Runs Out Fast, and Musk Promises Fixes. One subscriber said a $200 plan that once lasted a week of heavy use ran out in 14 hours. On Sunday, Elon Musk said fixes to make Grok Bot use less computing power were underway, but gave no date.
Three Top Researchers Say Today's AI Isn't Conscious and Probably Can't Be. Writing in Nature, Anil Seth, Ned Block and Giulio Tononi call the idea that the right computer program alone creates a mind "fatally flawed." Their short version: "They are smart tools, not smart people."
TRY THIS PROMPT (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🔧 Gadget Starter: Describe a gadget you wish existed and get a real parts list and weekend plan
Build a single-file HTML app called Gadget Starter in vanilla HTML, CSS and JavaScript. No API key needed.
I describe a small gadget I wish existed (like a light that turns green when my plant needs water) or tap one of four example buttons, then pick my experience level and a budget.
The app calls the Anthropic Messages API with the web_search tool to find real beginner parts sold today, like starter boards, sensors and LEDs, with current prices and stores. It must never invent parts, prices or links, says "unknown" when it can't confirm a price, and only suggests low-voltage USB or battery builds.
Show the gadget's name, a difficulty badge, an estimated total, a parts table, a five-step first-weekend plan, a no-solder option, safety notes, a free simulator to practice in, and real source links.
Style: near-black background, mint green and amber accents, Space Grotesk, Newsreader and JetBrains Mono fonts.What this does: Describe a gadget you'd love to have, like a beginner-sized version of today's lead. The app searches for real starter parts and current prices, then gives you a parts table, a rough total, a five-step plan for your first weekend and safety notes, with links to where it found everything.
WHERE WE STAND(based on today's news)
✅ AI Can Now: Design the circuit board, case and software for a small working gadget, with files a factory can build as is.
❌ Still Can't: Build or safety-test it. Diode's engineers assembled and tested all three, and none has had durability or certification checks.
✅ AI Can Now: Turn a plain-English idea into a working map app with routes and local search, in Google's AI Studio.
❌ Still Can't: Reliably finish real app changes end to end. The best agent in Alibaba's test got the whole job right 26.4% of the time.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
🎬 General Magic (2018) - Movie
This 92-minute documentary follows the Apple spinoff that shipped a handheld, smartphone-like computer in 1994, years before anyone was ready for it. The product flopped, but the people behind it went on to shape the phones in our pockets. It's a great reminder that a working prototype is the easy part, which is worth remembering while AI starts designing gadgets.
How was today's edition? Tap one below, then tell me what you'd change on the next screen. I read every answer.
How was today's edition?
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.
