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
OpenAI Disables 4 ChatGPT Tools to Block Attacks. It's Not Enough
TLDR: OpenAI just added a new security setting to ChatGPT called Lockdown Mode, but the company admits it still can't fully stop attackers from stealing your data.
The Story:
On June 6, OpenAI launched Lockdown Mode, a free optional setting now available to all ChatGPT users including Free, Plus, Pro, and Business accounts. Turning it on shuts down live web browsing, deep research, agent mode, and the ability to pull images from the web. The goal is blocking "prompt injection" attacks, where hidden instructions in a website or uploaded file trick the AI into leaking your private data. But OpenAI admits ChatGPT can still be tricked even when Lockdown Mode is on. The company calls prompt injection a hard research problem it hasn't solved, and says this setting is mainly meant for people handling sensitive work data, not everyday users.
Its Significance:
Prompt injection has been a known weakness in AI chatbots since GPT-3 came out years ago. The fact that OpenAI is shipping a feature that turns off half its own product to barely fix the problem tells you how serious it is. If you've been thinking about hooking ChatGPT up to your most sensitive files and financial life, this is your sign to pump the brakes and consider where the technology is in terms of security. For most people, treating AI as a smart assistant you ask questions and connect to things that improve productivity, not a tool with the keys to everything you own, is still the safer call right now.
QUICK TAKES
The story: Researchers used Anthropic's Claude Opus 4.8 to find a serious bug in the privacy coin Zcash that had been hiding in the code for over four years. The bug would have let an attacker make unlimited fake Zcash coins, and the price of ZEC crashed 38% after the news came out.
Your takeaway: AI tools are now finding software bugs that human experts missed for years. That's great when researchers use them, but anyone with access to a strong model can run the same kind of search. Expect to see a lot more bugs surface in the next few months.
The story: Microsoft researchers found a bug in Anthropic's Claude Code GitHub Action that let attackers hide bad instructions in GitHub issues or pull requests. When Claude reviewed them, it could be tricked into leaking passwords and API keys out of the codebase. Anthropic patched the bug in May after Microsoft reported it.
Your takeaway: This is the second prompt injection story in today's issue for a reason. Coding agents that read messages from strangers and have access to your secrets are still a security problem nobody has fully fixed.
The story: A new interactive walkthrough breaks down how large language models actually solve arithmetic problems. The answer surprises a lot of people. Models don't use real math algorithms. They use what researchers call a "bag of heuristics," basically a pile of pattern-matching shortcuts.
Your takeaway: This is part of why ChatGPT confidently gets simple math wrong. It isn't computing, it's guessing based on patterns it's seen before. Use a calculator for anything that needs to be exact, especially financial numbers.
TOOLS ON OUR RADAR
💳 Wallos Free and Open Source: A useful personal financial utility that meticulously tracks all of your monthly internet subscriptions and reminds you exactly when your upcoming payments are due so you never waste money again.
📓 TiddlyWiki Free and Open Source: A unique and incredibly powerful personal notebook contained entirely inside one single file that you can easily carry on a portable drive and open with absolutely any standard web browser.
✍️ Manuskript Free and Open Source: A brilliant writing assistant built specifically for aspiring novelists that helps you meticulously outline complex fictional worlds develop interesting characters and carefully organize your massive story chapters.
🕵️ Mullvad Free and Open Source: A fiercely private internet browser created in partnership with legendary security researchers that specifically aims to completely hide your unique digital footprint and flawlessly block invasive tracking scripts out of the box.
TRENDING
Meta Is Building Data Centers That Are Just Tents Filled With AI Chips - Meta put up six giant weatherproof tents in Ohio, each 125,000 square feet, stuffed with billions of dollars in AI chips and powered by gas turbines on-site. The tents take months to build instead of the two to three years a real data center needs.
AI Has Come for Serif Fonts - AI companies like Claude and Perplexity are switching to old-fashioned serif fonts to look more human and trustworthy. Critics have started calling the trend "tasteslop."
I Design With Claude Code More Than Figma Now - A Jane Street designer says they've mostly stopped making Figma mockups. Instead, they write a short description, and Claude builds working prototypes directly in the real codebase, which they test for days before pitching.
Banks Lay Groundwork for Mass Workforce Cuts as AI Takes Hold - Major global banks are openly preparing for big AI-driven layoffs in back-office and operations roles. Bloomberg Intelligence has projected up to 200,000 banking jobs could be cut over the next three to five years.
Perplexity's Search as Code Lets AI Models Write Their Own Search Pipelines - Instead of calling a fixed search API, Perplexity now lets AI agents write Python code to control how the search runs step by step. Early tests show 2.5x better results on complex research tasks.
Nvidia's Nemotron 3 Ultra Becomes the Smartest Open US Model, but China Still Leads - Nvidia's new 550-billion-parameter open model scored 48 on the Artificial Analysis intelligence ranking, the highest of any open American model. China's Kimi K2.6 still leads the open-weight field at 54.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
Describe your business or job. Get the hidden income streams you're sitting on but not monetizing, with realistic dollar estimates.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Untapped Revenue Audit — find hidden income streams in what someone has already built. Persist to localStorage key 'untapped_revenue_v1'.
Aesthetic: deep money green (#0c1410), soft green/gold radial gradients. Inter sans 800 for headings, Source Serif 4 for body italic, JetBrains Mono for labels and dollar amounts. Mint green (#50c884) primary with glow effects on the big number. Amber (#dca050) for medium-effort tags, soft red (#dc7850) for the blocker card. Big bold dollar potential number with text-shadow glow.
Form: business textarea, assets textarea (with strong help text listing audience/content/IP/network/equipment), current revenue dropdown (under $50k / $50–150k / etc), time available dropdown, hesitated-to-charge-for textarea.
System instructions to the model: act as revenue strategist who mines existing assets — don't suggest random new businesses. Be specific to what they wrote. Use real revenue numbers based on their current scale. Honest about effort vs payoff. Return raw JSON: total_potential (conservative $ for year 1, formatted like '$35K'), total_summary (1-2 italic sentences), streams array (4-6 sorted fastest-first: name + revenue + period + effort low/med/high + speed fast/med/slow + pitch 1-2 sentences + who_pays + what_to_sell + first_step concrete action this week), quick_wins array (3-4 items: text + $ amount), biggest_blocker (2-3 sentences naming the real reason — pricing fear/identity/perfectionism/distribution).
Render: giant total-potential headline number in green with glow on dark gradient background. Streams wrapped in a list with each card showing numbered stream / name / colored effort + speed tags / large revenue number / italic pitch with green border / two-column who-pays + what-to-sell / green-bordered "first move this week" card. Quick wins as green-tinted rows with number + text + dollar amount. Red-bordered "biggest blocker" card with italic honest text. Archive shows total potential per audit.What this does: Tell it what you do, the assets you've already built (audience, content, IP, expertise, network), your current revenue range, time available, and what you keep giving away or hesitating to charge for. Get back a giant headline number for realistic untapped potential, 4–6 specific revenue streams sorted by speed-to-revenue (each with annual revenue estimate, effort and speed tags, pitch, who pays, what to sell, and one concrete first move for this week), 3–4 quick wins bankable in 30 days with dollar amounts, and an honest naming of the real reason you haven't already started. Saves to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Find serious software bugs that hid in code reviewed by human experts for over four years (Zcash, Claude Code GitHub Action)
❌ Still Can't: Fully block prompt injection attacks. Even OpenAI's new Lockdown Mode admits it can't stop them.
✅ AI Can Now: Build working web prototypes from a short text description, replacing some traditional design steps
❌ Still Can't: Do real arithmetic. Models guess answers using pattern shortcuts, not actual math.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A collection of short stories from the author of Revelation Space, including the title story about a famous artist whose paintings get bigger and simpler as he ages, and "Beyond the Aquila Rift," about a deep-space cargo pilot who wakes from cryo-sleep somewhere unfamiliar. Reynolds writes tight AI puzzles with the lightness of an Aesop's fable. Two of the stories were adapted in Netflix's Love, Death & Robots, but the collection itself is barely known.
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.
While the Dollar Weakens, Gold Doesn't Ask Permission.
Physical gold sits outside inflation, banking risk, and government policy and a Gold IRA lets you own it with the same tax advantages you already have. True Gold Republic's free 2026 kit shows you exactly how.