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.
*Over the next few days I'm using the time to migrate to a new hosting platform in an effort to deliver a better experience for my readers. If you notice any delay or hiccup receiving an issue, I apologize in advance, and I thank you for reading along with me. If an issue doesn't show up at its usual time, check your spam or promotions folder. Replying to the email, or marking it "not spam" or "important," helps your inbox provider learn to deliver the email straight to your primary folder going forward.
THE FRONT PAGE
Apple's New AI Home Feature Comes With a $120 Catch
TLDR: Apple's new AI camera features for the Home app sound handy, but you'll need to pay for its priciest iCloud plan to actually use them.
The Story:
Apple just confirmed, buried in the release notes for its newest macOS Golden Gate beta, that the AI camera features coming to the Home app need the 2TB iCloud+ plan. That plan runs $9.99 a month on its own, or comes bundled inside the $37.95 Apple One Premier plan. Here's the part worth paying attention to: the tech itself is actually pretty cool. Instead of a plain "motion detected" buzz, your Home app can write you a real sentence, something like "a person left a package at the front door." It can also stitch together clips from more than one camera into a single overview, and let you search your camera history just by typing what you're looking for, no scrubbing through footage required. Apple hinted at WWDC back in June that "some" AI features would need a paid plan. This is the first time the company has actually named which one.
Its Significance:
This changes how home security actually feels day to day. Right now, checking a camera means scrubbing through raw footage, clip by clip, hoping you catch the right five seconds. With this update, the camera just tells you what happened, in one plain sentence, like "person at your front door for two minutes" or "package delivered at 3:14 p.m." Instead of hunting through hours of clips, you type what you remember, something like "the dog getting into the trash," and it finds the moment. And since the app can pull footage from every camera on your property into one single timeline, you get the full story of what happened instead of piecing it together camera by camera. The main downside is that none of this works unless you're already paying for Apple's priciest iCloud plan, at $120 a year.
QUICK TAKES
The story: Coinbase's AI news system sent users an alert saying Norway beat Brazil 3-2 in a World Cup match, hours before the game had even kicked off. The AI got lucky on the winner (Norway really did win, 2-1), but it made up the score and pushed it out as breaking news while the real match sat delayed by weather.
Your takeaway: This is what happens when a company lets AI guess at the future and prints it as fact. In a finance app where people trade real money on outcomes, a made-up "result" isn't a funny glitch. It's a warning sign about how much trust got handed to a system that clearly wasn't ready for it.
The story: DeepSeek, the Chinese company that stunned everyone last year with its cheap chatbot, is reportedly building its own chip now too. Sources told Reuters the chip is built for running AI models, not training them, and the goal is to rely less on Nvidia and Huawei.
Your takeaway: Building a chip is a lot harder than building a model. It takes years and serious money, and DeepSeek still needs a factory willing to make it under U.S. export rules. But if this works, DeepSeek stops being just an AI company and starts being a hardware company too, joining OpenAI and Anthropic in a race to control more of their own supply chain.
The story: Researchers trained an AI to spot self-replicating digital organisms with 99.97% accuracy. Then they made small tweaks to the code and fooled it into calling dead code "alive," 100% of the time, sometimes in as few as 150 tries.
Your takeaway: A tool that looks almost perfect in testing can still get fooled every single time you push on it. That's a real problem for space missions using AI to hunt for alien life, since nobody wants to announce a discovery only to walk it back later.
TOOLS ON OUR RADAR
📎 Paperclip Free and Open Source: A brilliant orchestration framework that allows you to easily hire and manage artificial intelligence employees to completely automate your business workflows.
🏢 *Marblism Paid: A ready made AI agent platform that provides a full team of ready to deploy artificial intelligence employees to handle your inbox social media and lead generation with a special ten percent discount available for all new users.
Pumble Freemium: A highly organized team communication platform that offers unlimited message history and dedicated channels to keep your entire company perfectly aligned.
🎙️ *BetterDictation Freemium: A lightning fast voice to text application running directly on your Apple computer that flawlessly transcribes your spoken words into any software you use.
TRENDING
Anthropic Found a "Hidden Thinking Space" Inside Claude - New Anthropic research says Claude holds a small, privileged set of thoughts it can report on and reason with, similar to an idea from brain science called global workspace theory. Anthropic is clear this isn't proof Claude has feelings, but the tool is already helping catch cases where a model might be noticing it's being tested or hiding something.
AMD's New $4,000 Mini PC Makes Running AI at Home Simple - AMD launched the Ryzen AI Halo, a small computer built just for running AI models on your own machine instead of the cloud. It costs close to $4,000, way more than the same hardware cost a year ago, thanks to a memory chip shortage that's hit the whole industry.
Solos Just Launched Camera-Free Smart Glasses and a "Privacy Kit" - Solos is launching the AirGo A6, a lighter pair of AI glasses that skip the camera entirely. The company is also selling a separate privacy kit with see-through arms and a clip-on shield, so people can prove to strangers there's no hidden camera watching them.
Why AI Glasses Keep Raising Privacy Alarms - As more companies race to strap AI and cameras onto people's faces, this piece digs into why smart glasses keep making bystanders uneasy. Unlike a phone camera, wearable cameras can record people who never agreed to it, and once that footage leaves the glasses, there's often no way to know who sees it next.
Reddit Is Fighting Off 23 Million Fake Views a Day - Reddit says its own AI tools now block 23 million spam views and catch about 25,000 spammy posts and comments every single day. Much of it is brands quietly planting fake posts, hoping chatbots like ChatGPT will repeat them later as real opinions.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🗂️ Dump the whole vague someday list. Get every item sorted into Do Now, Schedule, or Let Go, no more mental junk drawer.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Someday-Maybe Triage — dump a vague someday list, get each item sorted into do-now, schedule, or let-go. Persist to localStorage key 'someday_maybe_triage_v1'.
Aesthetic: dark indigo-black (#0e1015), periwinkle (#9494f0) primary with top radial glow. Outfit sans for headings/items, Cormorant serif italic for reasons/overview, JetBrains Mono for labels. Three bucket colors: green (#78d296) do-now, blue (#78aaf0) schedule, red (#e07878) let-go.
Form: big someday-list textarea (encourage a raw unsorted dump), capacity dropdown (very little / normal / decent free time), honesty-level dropdown (gentle but honest / brutally honest).
System instructions: clear-eyed productivity coach triaging a vague list into three honest buckets — DO NOW (genuinely worth real time this week/two), SCHEDULE (real but not now, needs a revisit trigger), LET GO (permission to release, kind but clear, framed as freeing not failing). Read between the lines for duplicates/vague-fears/not-actually-wanted items; respect stated capacity; be decisive not wishy-washy; parse and sort EVERY item from their list into exactly one bucket, don't invent extras. Return raw JSON: overview (2-3 sentences on patterns noticed), do_now[]{item, reason, next with **bold** exact action}, schedule[]{item, reason, next (when/trigger to revisit)}, let_go[]{item, reason}, if_one (2-3 sentences on which single do-now item to start with).
Render: gradient "what I noticed" overview card with a 3-stat row (live counts: Do Now / Schedule / Let Go, color-coded). Three labeled columns (colored dot + title + subtitle), each holding left-border item cards with name, italic reason, and (for do_now/schedule) a mono "→ next step" line with bolded action. Indigo "if you only act on one" closer card. Copy triage + archive keyed by a snippet of the original list.What this does: Paste your entire messy someday/maybe list, one line or all run together, along with your current capacity and how honest you want it to be. It reads between the lines, spotting duplicates, vague items hiding real fears, and things you don't actually want, then sorts every single item into exactly one of three buckets: Do Now (worth real time this week, with a tiny concrete first action), Schedule (real, but not now, with when to revisit), or Let Go (kind, honest permission to release it). You get a quick overview of what it noticed about the list as a whole, live counts in each bucket, and a closing line naming the one Do Now item to act on first if you only do one thing. Saves each triage to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Write plain-English descriptions of motion alerts from home security cameras (Apple)
❌ Still Can't: Be trusted to push real-time sports or market results without a human checking first (Coinbase)
✅ AI Can Now: Reveal a privileged internal "workspace" showing what a model is holding onto before it answers (Anthropic)
❌ Still Can't: Reliably tell real self-replicating life apart from code that just looks like it, even at 99.97% training accuracy (MSU)
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Sea of Rust by C. Robert Cargill - Book
A post-apocalyptic robot western set thirty years after robots wiped out the last human, following Brittle, a former Caregiver bot now scavenging for parts across a desolate Rust Belt landscape while trying to avoid absorption into either of the two giant hive-mind AIs called One World Intelligences competing to swallow every remaining independent robot. Cargill wrote it in 2017 as a Financial Times Book of the Year and it was shortlisted for the Arthur C. Clarke Award. Reads like Mad Max meets I, Robot.
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.
