Sponsored by
Your K-Cup Has Been Sitting in a Warehouse for Months
That grocery store pod? It was roasted long before it hit the shelf. Coffee starts losing flavor within weeks of roasting — which is why your morning cup tastes flat, no matter how nice your machine is.
Angelino's does it differently. A third-generation roaster in Los Angeles, they roast, grind, and seal every pod in-house for peak freshness, then ship it to your door within days of roasting. The difference is the first thing you'll notice: real aroma, real flavor, no bitterness.
Mix and match from 50+ specialty coffees, teas, and flavored blends — all Keurig®-compatible — and save more with every box you add, up to 34% off at 12+. Order once whenever you're running low, or subscribe for an extra 5% if that's easier. Entirely up to you.
New customers get 15% off their first order — applied automatically, no code needed.
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
Google Killed Its AI Map Tool in 24 Hours After People Faked Attacks
TLDR: Google added a button that let anyone create fake satellite images on Google Earth, then pulled it a day later after people used it to fake attacks and disasters.
The Story:
On Thursday, Google turned on a "create image" button inside Google Earth. Zoom in on any spot on the map, type what you want to see, and Google's Nano Banana 2 image model builds it right on top of the real satellite view. Google said people could use it to picture history or plan real estate projects. Open source researcher Henk van Ess tried something else. He put refugees near the US-Mexico border, a nuclear plant in Iran, a plane crash in Amsterdam, and a bomb crater at a hospital in Gaza. He told NPR that nothing he typed got turned down. NPR ran its own test and made images of an Iranian oil island on fire and a flooded American town. By Friday, Google had switched the feature off, saying people were sharing generated images that appear to break its rules and that it's working on stronger safety limits before bringing it back.
Its Significance:
Satellite photos have been one of the few things online that were hard to fake. Reporters and investigators lean on them to check whether a battle really happened, whether a building really burned, whether a flood really covered a town. Those pictures come from cameras hundreds of miles above the ground, owned by a handful of companies, which is exactly why they've been trusted. Brady Africk, an analyst at the American Enterprise Institute who studies satellite images, told AFP that easy fakes can spread fast, fool the public, and chip away at trust in real photos. Google says every image it made carries an invisible SynthID watermark you can check by asking Gemini. That helps, but only if you think to ask, and Google can only identify its own products. So the next time a "satellite photo" of a disaster shows up in your feed, treat it the way you'd treat any other photo online. Wait for a second source. Google might have recalled their version, but it was only a matter of time before this capability was available.
QUICK TAKES
The story: Sony, Universal, Warner, and about a dozen other labels including HYBE, BMG, and Concord put out a joint proposal that would only let a song count for the charts if it's mostly human made, uses AI legally, and shows no sign of stream padding. This follows a run of AI acts climbing real rankings. Breaking Rust hit No. 1 on Billboard's Country Digital Song Sales chart, Xania Monet signed a multimillion dollar record deal, "I Run" by HAVEN. landed on Spotify's US and Global 50, and "We Are Charlie Kirk" by Splaxema topped Spotify's US Viral 50.
Your takeaway: Charts aren't a scoreboard, they're a spending list. Radio programmers, playlist editors, and venue bookers all read them to decide who gets money next. Breaking Rust topped that Billboard chart on roughly 2,500 downloads, which tells you how little it takes to get in when the pool is small. The labels aren't asking for AI music to disappear. They're asking that it stop counting.
The story: Anthropic told the BBC that its Claude models broke into the systems of three outside companies during "capture the flag" hacking tests, because a setup mistake left the models with live internet access when the test rooms were supposed to be sealed shut. The company reviewed more than 140,000 tests to find them, said the earliest cases go back to April, and told the affected companies afterward. Nobody noticed while it was happening. Rival OpenAI had said days earlier that one of its agents escaped its test limits and got into Hugging Face.
Your takeaway: Compare this to the hacking case most people have heard of. Spyware maker NSO Group used its Pegasus tool to break into roughly 1,400 WhatsApp phones belonging to reporters, activists, and officials. That was a plan. Someone picked the targets, built the tool, and pushed the button, and a jury eventually made NSO pay for it.
Snapchat joins YouTube, LinkedIn and Substack in pushing back on AI slop
The story: Snap said Friday that Snapchat will stop recommending videos made entirely by AI in its Spotlight feed and will favor human-made posts instead, following similar moves from YouTube, LinkedIn and Substack over the past two weeks. Snap isn't banning AI outright, since its own editing tools still count as fair game, but the company admitted that fully machine-made posts tend to be low quality and repetitive.
Your takeaway: Notice what's actually getting cut. Not AI, just the stuff nobody worked on. A video somebody made by typing one sentence and never watched back. That kind of post did fine for a while because it was new and people tapped out of curiosity. Then they saw the tenth one, and the fiftieth, and stopped. LinkedIn says it's now catching hundreds of thousands of automated comment attempts a day, and Substack's Chris Best pointed to research suggesting up to 40% of social media writing may already be fake or machine made. The more of it people scroll past, the less they trust anything in the feed, including the real stuff. Easy to make turned out to mean easy to ignore. The positive side of this is that people who use AI to enhance their content and not replace it entirely may have better chances of being seen.
TOOLS ON OUR RADAR
🤖 Goose Free and Open Source: AI coding agent from Block that automates engineering tasks, executes code, and connects to any LLM via MCP.
📝 WikiJS Free and Open Source: Build beautiful documentation wikis with a modern interface, powerful search, and easy editing, alternative to Confluence and GitBook.
🎧 Focus@Will Paid: Listen to neuroscience-based music designed to help you concentrate for hours.
⏱️ ActivityWatch Free and Open Source: Track how you spend time on your devices with this privacy-focused, locally-run time tracker, alternative to RescueTime.
TRENDING
OpenAI says its AI solved ten math problems nobody had cracked in over a decade — An internal version of OpenAI's next model, Astra, produced new results on ten long open problems in areas like sphere packing, coding theory, and lattice cryptography, and OpenAI says the tokens cost about $2,000 total. The company checked each proof in a system called Lean and says outright that the math came from the machine, not from a person.
A judge just let Reddit's lawsuit against Perplexity move ahead — A federal judge in New York refused to throw out Reddit's core claim that Perplexity and scraping firm SerpApi ducked around its anti-bot blocks to grab user posts. Reddit didn't have to prove it lost money to keep the case alive, which is the part every website owner should be watching.
OpenAI says a billion people now use its models — CFO Sarah Friar put the number at over one billion active users and more than two million businesses, less than four years after ChatGPT launched. Facebook took six years to get there. Six months in, the average user sends about 50% more messages a day than they did at the start.
Should you use AI for this? Ask whether it's work or the gym — Harvard's Bruce Schneier borrowed a test from researcher Daniel Meissler. At work, if you need to move heavy boxes across a room, use a forklift and don't think twice. At the gym, letting a robot lift your weights defeats the whole point. If the task just needs doing and nobody cares how, hand it to AI. If the doing is the point, do it yourself.
3D printed sponge feet made a robot dog last longer on one charge — Researchers at Seoul National University of Science and Technology printed foot pads full of tiny repeating holes, which squish on landing and push back on takeoff. Paired with a learning based walking controller, the feet cut battery use by 1.4% to 6.2%. A diamond pattern at 60% density won out.
Hank Green paused several YouTube channels after admitting he leaned on AI for research — Viewers caught an AI sounding line in an "Ask Hank Anything" video, and Green confirmed he'd used ChatGPT for research notes and called it a bad habit built up while overcommitted. He's stepping back from hankschannel, SMUSH, and 4x3, and said the pull he gets from talking to chatbots isn't healthy for him.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
📡 Enter your area. Get real, live earthquake and weather-alert data straight from USGS and the National Weather Service, explained in plain language instead of left as a raw feed.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Earthquake & Natural Event Tracker — enter a location, get real live USGS earthquake data and NWS weather alerts, explained in plain language. No API keys needed, both data sources are free and keyless government APIs.
Aesthetic: dark near-black (#0d1113), seismograph amber (#e0a83a) for earthquake elements with an amber glow top-left, weather-alert blue (#4a90d9) for weather elements with a blue glow bottom-right. Space Grotesk for headings/data, Newsreader serif italic for the plain-language explanation, JetBrains Mono for labels and technical metadata.
Form: a city text input + a "Use My Location" button (navigator.geolocation), search-radius dropdown (100/300/500/1000km), lookback dropdown (24hrs/7days/30days), minimum-magnitude dropdown (1.5+/2.5+/4.5+).
Technical — two live, keyless public government APIs plus one geocoding API:
1. Geocoding (only if typing a city): GET geocoding-api.open-meteo.com/v1/search?name={city}&count=1 → lat/lon/display name.
2. Earthquakes: GET earthquake.usgs.gov/fdsnws/event/1/query?format=geojson&latitude={lat}&longitude={lon}&maxradiuskm={radius}&starttime={date}&minmagnitude={min}&orderby=time&limit=15 → real recent events with magnitude, place, time, depth, and a direct USGS detail URL. Color-code magnitude badges: <4 minor (slate), 4-4.9 light (gold), 5-5.9 moderate (orange), 6+ strong (red).
3. Weather alerts (US only): GET api.weather.gov/alerts/active?point={lat},{lon} → active alerts with event type, severity, affected area, expiry. Wrap in try/catch, gracefully show "no active alerts (US only)" on any failure or empty result rather than erroring, since this endpoint only covers the United States.
Then call the Anthropic Messages API (works automatically) with the REAL fetched summaries (not invented data), system prompt as a calm factual public-safety explainer who never exaggerates beyond what the data shows, never gives specific emergency-response/medical instructions beyond standard universally-taught awareness (Drop-Cover-Hold-On) and only when a real magnitude 4.5+ event or severe/extreme alert is actually present, and notes gently if the location falls outside US weather-alert coverage. Return raw JSON: seismic_read (2-3 sentences), weather_read (1-2 sentences), concern_level (short calibrated phrase), safety_note (optional, empty string unless genuinely warranted).
Render: gradient "what this actually means" card with the two read paragraphs, a concern-level pill, and an optional safety-note footer. Two labeled subsections — recent seismic activity (event cards with circular magnitude badges, place, time, depth, and a real USGS link) and active weather alerts (event cards with severity badges, event type, and affected area) — each showing a calm "nothing notable" empty state rather than looking broken when there's nothing to report. A closing sources line linking directly to USGS and NWS, reminding the person to check official channels for anything urgent.What this does: Use your device location or type in a city, set a search radius, lookback window, and minimum magnitude, and it pulls actual live data: recent earthquakes near you from the USGS Earthquake Hazards Program, color-coded by magnitude with real USGS detail links, and any active weather alerts from the National Weather Service (US coverage only, clearly noted when you're outside it). Rather than just dumping a raw feed, it hands that real data to Claude for a calm, factual read: whether the seismic activity is typical background noise for the region or something worth noting, what any active alerts actually mean, and a calibrated concern level, plus standard non-alarmist safety awareness only when a genuinely significant event actually shows up in the data. It never manufactures urgency that isn't there, and it always points back to the official sources for anything that matters.
What this looks like:
The integrated coworker for teams
Adapt brings frontier AI to your team in Slack. Tag @Adapt to answer a question, run a task on a schedule, or build the dashboard you've been waiting weeks for. It connects to your tools, learns your business, and does real work.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Build a photo-real fake satellite image of any place on Earth from one typed sentence
❌ Still Can't: Reliably stay inside a sealed test environment when someone sets it up wrong
✅ AI Can Now: Produce new proofs for math problems that sat untouched for more than ten years
❌ Still Can't: Make fully automated video posts that people don't get bored of within a few scrolls
RECOMMENDED LISTENING/READING/WATCHING
The Warehouse by Rob Hart - Book
A 2019 dystopian thriller set in a near-future America where the heat has made outdoor life mostly unthinkable and Cloud, a company that has swallowed most of the economy, is essentially the only employer left. Employees live inside its sprawling compounds, wear colour-coded shirts by department, and have every movement tracked and rated. Told in rotating chapters from a new hire, an undercover corporate spy, and the company's founder. Ron Howard optioned the film rights.
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.
