Sponsored by
Moda is the AI design agent with taste
Moda's viral launch hit 4.4 million views in two days. Tens of thousands of professionals signed up. Startups, agencies, forward-thinking brands and top firms are now using Moda to create brand-aligned slides, ad creative, reports, social carousels and more.
Most AI tools tend to create what we call "AI slop": repetitions of the same colors, layouts and fonts. And when you try to fix it, you get stuck in a loop of re-prompting.
Moda is different. Drop in your website URL, and Moda learns your brand from the ground up: your colors, your fonts, your visual language. Then it helps you generate pro-quality slides, docs, and marketing assets.
The best part? Every layer is fully editable on a real canvas, and exports to powerpoint, PDF and more.
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
SpaceX Just Paid $60 Billion for Cursor, the AI That Writes Code
TLDR: SpaceX is buying Cursor maker Anysphere for $60 billion, pulling rockets, satellites, a chatbot, and a top coding tool under one roof.
The Story:
SpaceX agreed to buy Anysphere, the company behind the popular AI coding tool Cursor, for $60 billion in an all-stock deal, according to Reuters. The news landed just days after SpaceX went public on the Nasdaq at a value of more than $2 trillion, making it one of the most valuable companies in the world. Cursor lets people write software by chatting with an AI, and its business has grown fast since 2022, pulling in around $2.6 billion a year from company customers. SpaceX already merged with Elon Musk's AI firm xAI, the maker of the Grok chatbot, back in February. Buying Cursor gives that AI side a real foothold in coding, where rivals like OpenAI and Anthropic got an early lead. The deal should close in the third quarter of 2026.
Its Significance:
This deal is about more than coding. SpaceX now owns rockets, satellites, a chatbot, and one of the most-used coding tools, all in one company. The thread tying them together is power and compute. SpaceX has even asked U.S. regulators for permission to launch solar-powered satellites that would run AI in orbit, soaking up sunlight instead of draining power grids and water down here. If that pans out, the same company that flies the rockets could also build the AI, write the code, and beam the energy down from space.
QUICK TAKES
The story: A study by SemiAnalysis found that if someone used a $200-a-month ChatGPT Pro plan all the way to its limit, the real compute cost could reach $14,000 at standard rates. The same report says OpenAI starts losing money on some plans once heavy users pass about 11% of the cap, while Anthropic holds out a bit longer, near 20%.
Your takeaway: Flat monthly AI pricing may not last. As more people run AI agents that work for hours at a time, expect usage caps, higher prices, or pay-as-you-go billing to spread.
The story: Alibaba rolled out the Qwen Robot Suite, its first set of AI models made for robots instead of chat windows. The models split a robot's brain into three jobs: finding its way around, guessing what happens next in a scene, and moving its body to act.
Your takeaway: China is pushing AI off the screen and into machines that can pick things up and move around. Robots that see and act are getting closer to warehouses, factory floors, and maybe kitchens.
The story: CrowdStrike launched Continuous Identity for AI Agents, which checks every single action an agent takes instead of trusting it once at login. It looks at who owns the agent, who's running it, and the risk right then, then grants or blocks access on the spot.
Your takeaway: AI agents move fast and can reach sensitive data, so a one-time "okay" isn't safe. Security tools that watch agents action by action are headed for the workplace.
TOOLS ON OUR RADAR
🌐 Kiwix Free and Open Source: An essential educational tool that allows you to download entire websites like Wikipedia or TED Talks so you can browse them completely offline anywhere in the world without an internet connection.
💸 Wallos Free and Open Source: A specialized personal finance application that helps you track all of your recurring subscriptions and payments in one place so you can identify unnecessary expenses and manage your monthly budget more effectively.
🔍 Changedetection.io Paid: A clever web monitoring tool that automatically notifies you whenever a specific website changes its content allowing you to track price drops restocked items or news updates without manually checking the page.
🖼️ Lychee Free and Open Source: A sleek and self hosted photo management system that allows you to upload and organize your digital images into beautiful web galleries that you can share privately with friends and family.
TRENDING
A Nuclear Trick Could Cool AI Data Centers With Zero Water - A startup called Ferveret, started by two MIT researchers, borrows a method from nuclear reactors to cool AI chips by dunking servers in a special liquid that makes tiny, fast-moving bubbles. In tests it improved power efficiency by 15% and used no water at all.
Facebook's AI Search Pulls Answers From Your Neighbors' Posts - Meta's new AI Mode answers questions using public Facebook posts, Groups, and Reels instead of plain web links. Handy for local tips, but answers built from random user posts can be wrong or out of date, the same worry people raised about Google pulling from Reddit.
Rio's Record-Breaking AI Model Turned Out to Be a Copy - Rio de Janeiro's IT agency released a 397-billion-parameter model that claimed to beat DeepSeek, but AI firm Nex showed the math: it's roughly 60% Nex's model and 40% Alibaba's Qwen stitched together. Without a custom prompt, the model even called itself "Nex" 79% of the time. The agency blamed an "incorrect upload."
Facebook Adds a Pile of New AI Tools - Meta is rolling out AI features across Facebook, including an "AI Mode" search tab, new photo and video editing tricks, and presets that can change the clothes or hair in your pictures. The creative bits are opt-in and can be switched off anytime.
A Frontier Model Isn't Enough on Its Own - A widely shared essay argues the AI race shouldn't end with a few labs owning all the value. The pitch: every company should build and own its own AI know-how, so the gains spread across industries instead of getting soaked up at the top.
13 Words on Reddit Can Fool AI Search - Cornell researchers found that a snippet of text as short as 13 words, planted on sites like Reddit or Wikipedia, can steer what AI search tools say. Since these tools cite user posts in about half of all answers, some brands are already gaming the system to push their products.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🏋️ Train like a memory champion. Real digit-span, n-back, card, and word drills that level up as you improve, plus a personalized regimen.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Memory Athlete Trainer — playable brain-training drills with adaptive difficulty plus an AI-generated 7-day regimen. Persist bests/sessions to localStorage key 'memory_athlete_v1'.
Aesthetic: dark forest-black (#0a0f0d), top mint radial glow. Inter sans 800 for headings, Newsreader serif for prompts/italic, JetBrains Mono for labels and the big stimulus numbers. Mint green (#3cdca0) primary, soft red (#f06478) for wrong feedback. Drill-selector tiles with emoji, big glowing stimulus display, n-back 3x3 grid with lit cells.
Four real interactive drills (vanilla JS, no AI needed for gameplay):
1. Digit Span — flash a number sequence, hide it, type it back; level = sequence length, auto-increments on success.
2. N-Back — 3x3 grid, squares light up one at a time every 2s over 20 trials; player hits MATCH when current position equals N-steps-back; score hits/misses/false-alarms; need 75%+ accuracy and ≤2 false alarms to advance N.
3. Card Sprint — flash N playing cards (rank+suit), recall in order with lenient suit-letter parsing (S/H/D/C → ♠♥♦♣); level = card count.
4. Word List — flash N words, recall in order; level = word count.
Each tracks personal best, shows current level + best, supports retry/switch, records session count.
One AI feature: "Get my personalized regimen" button calls the model with the user's current bests. System instructions: act as elite memory-sports coach, design progressive 7-day plan across the 4 drills, reference real techniques (chunking digits in 3-4s, PAO for cards, Major System for numbers, spaced practice), include one recovery day. Return raw JSON: assessment (2-3 sentences referencing their actual numbers), regimen array (exactly 7 days: day + focus with **bold** drill names and specific targets).
Render: header with session counter. Drill-tile grid (4 tiles with emoji/name/desc/best). Arena card with title + level/best stats + dynamic body that swaps per drill state (memorize → recall → result). Personalized-regimen button below that reveals gradient assessment card + 7-day plan with day-number pills and bolded targets. Personal-bests grid at bottom. All gameplay client-side; only the regimen uses the API.What this does: Four playable brain-training drills with adaptive difficulty: Digit Span (memorize growing number sequences), N-Back (hit MATCH when a square repeats its position N steps back, with hit/false-alarm scoring), Card Sprint (memorize a sequence of playing cards and recall them in order), and Word List (ordered word recall). Each drill auto-levels-up when you succeed and tracks your personal best. Then one AI button generates a structured progressive 7-day regimen based on your actual scores, referencing real techniques (chunking, PAO card systems, the Major System). Bests and session count persist in localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run a robot through a real room, helping it find an object, plan a path, and pick that object up (Alibaba's Qwen Robot Suite).
❌ Still Can't: Be trusted to act on its own without a check on each step. A 13-word note planted on Reddit can change what an AI search tool tells you.
✅ AI Can Now: Switch between reasoning out loud and thinking in hidden internal states to save effort, a method used in the Rio model.
❌ Still Can't: Prove on its own that it built its own skills. A model can be sold as original work and actually be two other models stitched together (Rio 3.5).
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Renaissance (2006) - Movie
A French animated sci-fi noir set in Paris in 2054, rendered entirely in stark black-and-white motion-capture, about a police detective investigating the kidnapping of a young scientist working on a classified project for a beauty and longevity corporation that runs the city. Daniel Craig voices the detective in the English dub. The animation alone is striking sci-fi imagery and the surveillance-state Paris is closer to where AI cities are actually heading than most live-action films get.
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.
8 levels of context maturity in AI-native engineering
More MCPs give agents access to information, not understanding. You agents are still missing the context to generate output you can fully trust. This 8-level model explains where teams get stuck and what fixes it. Join this live webinar on June 24 (FREE) and grab the free assessment to see where your team stands.