In partnership with
Your site is losing leads. You can't see where.
Your site looks fine to you — and it's still losing leads. SureThing renders any URL in a real browser, scores 9 dimensions, and ranks the fixes by wasted impact.
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
Runway's AI World Builder Draws Apps Frame by Frame
TLDR: Runway built an AI that generates a working app one video frame at a time, skipping code entirely, and in its own tests people picked it over a coded version 61% of the time.
The Story:
Runway announced Solaris on August 31. It calls the model the first in a family it named Interface World Models. Right now, every app you use starts as a visual design that somebody has to turn into code before it can do anything. Solaris drops that middle step. It paints each frame of the screen as you click and drag, in real time at 720p, with no HTML or JavaScript underneath. A language model decides what should happen next, and Solaris draws it.
Runway tested it two ways. First it asked top models, including Claude Fable 5, GPT-4o, and Gemini 2.5 Pro, to rebuild 30 web pages from a single screenshot, and every one of them lost detail along the way. Then it put Solaris head to head with Claude Opus 5 on 30 interaction tasks and asked 250 people to compare the results, collecting close to 7,500 judgments. Solaris won on following instructions 61% to 24%, and on behaving naturally in the scene 71% to 21%.
Its Significance:
Picture a store where you drag a shirt onto a photo of yourself, or a tutorial that redraws the next step using your own kitchen instead of a stock video. That's the pitch, and it's a different way of building software than anything most of us use today.
Runway is upfront about what's not there yet. Solaris struggles to draw text that stays still and readable, which is a strange weakness for something meant to render menus and buttons. It can produce a confident wrong answer with no way to trace it, and generated pixels don't work with screen readers, so a blind user won't get much value yet.
QUICK TAKES
The story: Google updated Teamwork in its Antigravity platform, a setup where groups of AI agents propose ideas, tear each other's work apart, and keep going for hours or days. Paired with Gemini 3.7 Flash, the agent teams cracked seven open research problems across math and computer science, hit 71% on TCSBench, and built a RISC-V CPU simulator that boots an operating system.
Your takeaway: One AI agent working alone tends to make a confident mistake early and then build everything on top of it. Putting several agents in a room to check each other is Google's fix, and it's the same reason your team does code review instead of trusting one person's first draft.
The story: On August 31, the European Commission classified ChatGPT as a Very Large Online Search Engine under the Digital Services Act, the first time an AI assistant has been put in that tier. OpenAI reported about 159.1 million monthly users in the EU, more than triple the 45 million line that triggers the rules, and it has until the end of December to comply.
Your takeaway: Europe now has real leverage over how ChatGPT works, and the price of ignoring it runs up to 6% of OpenAI's worldwide revenue. The list of new duties includes yearly risk reviews, data access for outside researchers, and a public record of every ad the platform runs, which arrives at an interesting moment for a company that just built a billion dollar ad business.
The story: The Dyson CameraJet puts a 100,000-pixel camera under the brush head, reads 28 images per second, and uses a model trained on 470,000 dental photos to find the gaps between your teeth. When it spots one, it fires a spray of rinse at it in about a tenth of a second. Six years of work, 661 engineers, 38 patents.
Your takeaway: Dyson says the camera only turns on when you press the button, and that no image is saved on the brush or in the cloud. No outside group has checked that, so you're taking the company at its word about a camera you put in your mouth.
TOOLS ON OUR RADAR
🌳 Forest Freemium: Grow a virtual tree while you stay off your phone, lose focus and it dies, part of the proceeds plant real trees.
📚 Calibre Free and Open Source: Organize, convert, and read your entire e-book collection across any format, works with virtually every e-reader.
🎤 Otter.ai Freemium: Get live transcription and AI meeting notes on Zoom, Teams, and Google Meet calls, with a genuinely free plan for light use.
📊 Airtable Freemium: Build a spreadsheet that acts like a database, with views, automations, and forms, no coding needed.
TRENDING
Florida Ordered AI License Plate Cameras Off Its Highways — The Florida Department of Transportation gave police 30 days to pull Flock and similar automatic plate readers off state roads, calling out misuse reports and privacy problems in the memo. It's the first statewide order of its kind, and Governor Ron DeSantis said he doesn't want Florida becoming a surveillance state. Privately owned cameras on business property aren't covered, and 214 cities have dropped Flock since 2021, with 90 of those cancellations in August alone.
OpenAI's Ad Business Hit $1 Billion a Year in Under 200 Days — ChatGPT Ads launched in February and is already on pace for $1 billion a year, with tens of thousands of advertisers and self-serve buying now open across India, Europe, the Middle East, and North Africa. Ads only appear for Free and Go users, and OpenAI says they never change your answers and advertisers never see your chats. Every rival with a big user base is going to run that same math this week.
Instagram Will Cut Your Reach If You Don't Say Your Influencer Is Fake — Instagram renamed its AI creator label to "AI-generated profile" and will limit how far unlabeled accounts spread if it decides the person in the posts isn't real. Add the label yourself and your reach stays the same. Using AI tools in your workflow doesn't count, so this is aimed squarely at accounts built around a person who doesn't exist.
OpenClaw Shipped Its Biggest Update Ever and Didn't Plan To — The open-source AI assistant went seven weeks without a release, which was odd for a project that had shipped 106 times in 230 days. What came out was version 2.0: 933 contributors, 569 of them first-timers, and more than 16,000 merged changes, about half of everything the project has ever merged. The team set out to fix installation and rebuild the browser app, then couldn't stop pulling the thread.
Quiq's New Tool Listens to Your Support Call and Acts Mid-Sentence — Voice Assist gives phone agents live suggestions that shift as the conversation moves, instead of the fixed scripts most agent-assist tools run on. It also does the task itself, sending a tracking link or a voucher on the same thread while the human stays on the line. Every suggestion passes the same verification step Quiq already uses on chat and text.
The Pentagon Added ChatGPT and Grok to Its Internal AI Portal — ChatGPT Mil and Grok for Government now sit alongside Google Gemini on GenAI.mil, the Defense Department's secure portal for its 3 million staff. More than 1.7 million people have already signed on. The military versions skip the data collection built into the consumer apps, and ChatGPT Mil is aimed at unclassified paperwork like logistics, planning, and policy.
A Harvard Law Dropout Raised $6M to Build an AI Rulebook for Cops — Blue Voice trains on a department's own laws, local rules, and protocols so an officer can look up the right rule in the moment instead of digging through a manual or guessing. Founder David Lawrence left law school after a campus shooting involving an officer and says general chatbots get these questions wrong up to 30% of the time. It's used daily by 225 county agencies across 25 states, and the tool points to the rule rather than telling the officer what to do.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, Grok, or Gemini)
✦Name the moment. Get the real sky that was actually overhead, rendered as a keepsake, not a guess.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Personal Star Map Moment — a date, time, and place, reconstructs the real night sky (bright stars + naked-eye planets) as a rendered keepsake.
Aesthetic: deep space near-black (#080a16), warm gold (#c8b478) primary with a gold glow top-left, soft indigo (#7888e0) secondary. Instrument Serif for the moment label and narrative caption, Inter for body/form, JetBrains Mono for labels and the chart's compass/planet-name text.
Form: a short personal label input, a place text input + "Here" geolocation button, a date picker (max today) and a time picker (default 9pm).
Technical — a from-scratch astronomical rendering pipeline:
1. A hardcoded 31-star catalog of J2000 RA/Dec for the brightest, most recognizable naked-eye stars, including a complete 7-star Big Dipper and 3-star Orion's belt tagged with asterism group + order for line-drawing, plus a brightness tier (1-3) for dot sizing.
2. Reuse the Kepler-solving orbital-mechanics functions (heliocentric position, Sun longitude, ecliptic→equatorial rotation) from a prior sky-guide build to compute real RA/Dec for Mercury through Saturn at the given moment.
3. Convert the local wall-clock date/time at the given place into a UTC instant using the geocoded location's IANA timezone (same offset-parsing approach as a prior jet-lag build), then Julian day, then Greenwich Mean Sidereal Time via GMST0 = (280.46061837 + 360.98564736629×d) mod 360, PLUS the observer's longitude for Local Sidereal Time — critically, since 'd' here is referenced to the orbital-elements epoch (JD 2451543.5) rather than the true J2000.0 epoch the GMST formula's constant assumes (JD 2451545.0), the constant must be epoch-corrected to 101.939089 (a 1.5-day / ~180° adjustment) — verify this isn't silently off by 12 hours using an independent check (Polaris' altitude must equal the observer's latitude regardless of time/date, since it's near the celestial pole) and a seasonal sanity check (summer evenings should show the Summer Triangle/Big Dipper high with Orion absent; winter evenings the reverse).
4. Transform RA/Dec + LST + latitude to real Alt/Az via standard spherical trigonometry for every star and planet; filter to alt>0.
5. Compute real moon phase/illumination via the synodic-month calculation.
Render a dome-style polar SVG sky chart (zenith at center, horizon at the edge, N/E/S/W labeled at the rim, two altitude reference rings): stars as dots sized by brightness tier, Big Dipper and Orion's belt connected with thin lines when their points are above horizon, planets as colored labeled dots (warm gold for Mercury/Venus/Mars, green for Jupiter/Saturn).
Then call the Anthropic Messages API (works automatically) with the real visible star names (top 6 by tier/altitude), real visible planet names, and the real moon phase. System prompt as a warm evocative keepsake-caption writer explicitly told to use ONLY the real objects given, never inventing constellations or objects not in the list. Return raw JSON: narrative (2-4 sentences).
Render: the moment label and place/date/time meta line, the sky chart, a legend, the moon-phase note, and a gradient narrative card with Claude's caption. A disclaimer noting this uses standard low-precision astronomical formulas as a decorative keepsake, not a precision planetarium.What this does: Give it a place, a date, a time, and a short label for the moment, and it reconstructs the genuine night sky from that exact instant: real bright-star positions from a 31-star catalog of the sky's most recognizable points (including a full Big Dipper and Orion's belt), real positions for Mercury through Saturn computed the same way professional low-precision ephemeris tools do, and the real moon phase for that night.
What this looks like:
AI Insights That Turn Your Data Into Bigger Profits
Your store is full of hidden growth opportunities. StoreClaw analyzes your Shopify and Amazon data, identifies the highest-impact actions, and helps you grow revenue while improving margins. Start free today with 10,000 bonus tokens. No credit card required.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Draw a working app interface one video frame at a time at 720p, responding to your clicks with no code underneath it
❌ Still Can't: Keep the text inside that interface readable and steady, which is a real problem for software made mostly of words
✅ AI Can Now: Read 28 camera frames per second inside your mouth and fire a jet of rinse at a gap between two teeth in about a tenth of a second
❌ Still Can't: Prove that camera isn't keeping anything, since Dyson's privacy claim rests on Dyson's word and nobody outside has tested it
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
eXistenZ (1999) - Movie
David Cronenberg's fleshy, paranoid vision of virtual reality gaming, about a legendary game designer on the run from assassins after the launch of her newest creation, a biological gaming pod that plugs directly into players' spines. Overshadowed on release by The Matrix, which came out the same year with a much bigger budget and a much cleaner aesthetic. Cronenberg's version is stranger, funnier, and more interested in what it would actually feel like to not be sure whether you're inside a simulation or not.
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.
