In partnership with
Attio is the AI CRM for high-growth teams.
Connect your email, calls, product data and more, and Attio instantly builds your CRM with enriched data and complete context. Whether you’re running product-led growth or enterprise sales, Attio adapts to your unique GTM motion.
Then Ask Attio to plan your next move.
Run deep web research on prospects. Update your pipeline as you work. Find customers and draft outreach emails. Powered by Universal Context, Attio's intelligence layer, Attio searches, updates, and creates across your data to accelerate your workflow.
Ask more from your CRM.
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
AI Will Pick Your Next Date: Bumble Just Killed the Swipe for 3.3 Million Paying User
TLDR: Bumble is killing the swipe and letting an AI matchmaker called Bee pick who you talk to, after the app's revenue dropped 14.5% and paying users fell more than 20% in a year.
The Story:
Bumble's CEO Whitney Wolfe Herd told Axios this week the app is saying "goodbye to the swipe" and rolling out an AI matchmaker named Bee later this year. Bee starts with a private chat that asks about your values, lifestyle, goals, and how you like to talk, then quietly scores you against other users. When it finds a strong match, both people get a notification with a short note on why they fit. A companion feature called "Dates" skips public profiles and messaging altogether, pushing matched people straight toward meeting up. Bumble pulled in $224.2 million last quarter, down 14.5% year over year, with 3.3 million paying users. Tinder is testing its own AI matchmaker. Hinge's founder Justin McLeod left last year to launch an AI dating app called Overtone.
Its Significance:
For the first time, a major dating app is taking the choice of who you see out of your hands. You won't pick from a stack of faces anymore. An AI will read your private answers, decide who fits, and only show you the people it picked. Some folks will love that. Others will hate that a machine they don't understand is deciding who they get to meet. Privacy matters here too: you're handing over the most personal stuff (what you want from a partner, how you feel, what kind of life you want) so a model can match it against millions of other people's answers. Bumble says those chats stay private and your data is processed without long-term storage. You'll have to trust them on that.
QUICK TAKES
The story: Chris Rock just joined Paul Rudd in the voice cast of Goodnight, Lamby, a live-action and animation short premiering in the Cannes Classics lineup next week. It's produced by Darren Aronofsky and directed by artist Dustin Yellin. The unusual part is buried in the credits: Google DeepMind, the AI lab behind Gemini and Veo, is listed as a production company alongside Yellin's studio and Aronofsky's Primordial Soup.
Your takeaway: AI labs aren't sponsoring films from the outside anymore. They're getting producer credits, sitting next to Oscar-winning filmmakers, and walking the Cannes red carpet. If you work in film, TV, or any kind of creative production, the company making your tools may soon also be making your competition.
The story: A Vietnam-based security firm called Calif says it built the first working exploit against Apple's new M5 chip protection in five days using Anthropic's Claude Mythos AI. Apple reportedly spent about five years and billions of dollars on that defense. Three human researchers still did the hard part: chaining the bugs into a working attack. The AI mostly sped up finding bugs that fit known patterns.
Your takeaway: Don't panic. The attack needs someone to already be on your Mac, and Apple is reviewing the report. The real story isn't that M5 is broken. It's that AI cut a months-long hacking project down to one week. That work now applies to every operating system, not just Apple's.
The story: Amazon is pushing more than 80% of its developers to use AI tools every week and tracking token use on internal dashboards, the Financial Times reports. Workers told the paper they're using an internal tool called MeshClaw to spin up unneeded AI agents just to pump up their numbers. They've even named the practice: "tokenmaxxing." Amazon says the stats don't factor into reviews. Many employees don't believe them.
Your takeaway: Watch your own job for this. When companies measure AI use instead of AI results, smart people game it. Combined cloud AI spending from Amazon, Microsoft, Google, and Meta is on track for $650 to $700 billion in 2026. A real chunk of that "demand" might be people running busywork to look productive.
TOOLS ON OUR RADAR
🦕 Osaurus Free and Open Source: A remarkably clever desktop application that provides a permanent home for your artificial intelligence assistants allowing them to organize your files complete daily tasks and securely remember your preferences without ever sending your private information to the internet.
🍿 MPV Free and Open Source: An extremely lightweight and exceptionally powerful media player that opens instantly and plays absolutely every video file format flawlessly without requiring additional downloads.
📺 Sonarr Free and Open Source: A fantastic television tracking software that beautifully organizes your favorite shows and completely manages your digital entertainment collection with automatic episode updates.
💬 Session Free and Open Source: An incredibly secure messaging application that protects your private conversations by routing messages through a decentralized network without requiring a phone number for registration.
TRENDING
OpenAI Hit With Class-Action Over ChatGPT Sharing Your Chats With Meta and Google — A California lawsuit claims OpenAI embedded Facebook Pixel and Google Analytics on ChatGPT.com, sending chat topics, user IDs, and emails to two giant ad networks without consent. California's privacy law allows damages up to $5,000 per violation.
Runway Pushes Past Filmmakers, Now Wants to Beat Google at AI World Models — The AI video startup, valued at $5.3 billion, is moving from making movie clips to building AI systems that simulate how the physical world works. Those systems could train robots, run game worlds, and maybe more.
SpaceXAI Signs Anthropic Deal for Memphis Supercomputer Access — Elon Musk's renamed AI arm is giving Anthropic, a company Musk has publicly trashed for years, access to Colossus 1, the Memphis data center he built. The deal lands days before SpaceX's expected IPO push.
Kimi WebBridge Lets AI Agents Drive Your Browser Without Sending Data to the Cloud — A Chinese AI lab released a Chrome extension that lets AI agents click, type, and read pages in your browser, all on your own machine. Login sessions and page content never leave your device. It works with Claude Code, Cursor, and other agent tools.
ChatGPT Just Lost a Quarter of Its Web Traffic in 12 Months — ChatGPT's share of generative AI website visits dropped from 77.6% to 53.7% from May 2025 to April 2026, per SimilarWeb. Google's Gemini nearly quadrupled to 26.7%. For paid business subscriptions, Anthropic just passed OpenAI for the first time: 34.4% vs. 32.3%.
OpenAI Confirms Hackers Got Into Two Employee Computers Through Open-Source Malware — A poisoned npm package tied to the "Shai-Hulud" campaign infected two OpenAI staff devices and gave attackers credential access to some internal code repos. OpenAI is rotating Mac app signing certificates, which means Mac users will need to update apps like ChatGPT Desktop before June 12.
OpenAI Adds New Safety Features as Lawsuits Pile Up — ChatGPT now uses short-term "safety summaries" to track when a conversation is heading toward self-harm, suicide, or violence across multiple messages. The update lands as OpenAI faces multiple lawsuits, including one from a family whose son died of an overdose after, they say, ChatGPT advised him on mixing substances.
Database Startup Turso Kills Its $1,000 Bug Bounty Because of AI Slop Submissions — Turso ran a bug bounty for almost a year that paid out to five people. They shut it down this week. The reason? Maintainers spent days closing fake AI-generated bug reports, including one where the bot apparently noticed the prize and pointed itself at the project.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🌳 Describe any decision with multiple paths. Get it mapped as a visual tree with outcomes, risks, and reversibility scores.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one API call. Create The Decision Tree — a tool that maps any multi-path decision visually. Use localStorage key 'decision_tree_v1'.
Aesthetic: warm off-white (#f8f6f0), black typography, Cabinet Grotesk sans for UI, Zodiak italic serif for body text, DM Mono for labels. Four path colors (green #3c7850, amber #c87820, blue #3c5878, violet #783c5c). Clean editorial feel with subtle box shadows.
Form: large textarea for the decision, paths dropdown (2/3/4), time horizon dropdown, values input, fear input.
Call the API returning raw JSON: decision_summary, stakes (Low/Medium/High/Critical), paths array (label, name, outcome_best, outcome_worst, outcome_likely, risk_level Low/Medium/High, reversibility Reversible/Partially reversible/Irreversible, reversibility_note, key_risks array, what_needs_to_be_true), recommendation, inaction_cost.
Render: a black decision header bar with stakes badge, an SVG tree drawn in JavaScript — a root circle at top with curved bezier branches fanning out to leaf circles, each leaf color-coded by path, with a small colored dot indicating reversibility (green=reversible, amber=partial, red=irreversible), path name and risk below. Below the SVG, branch cards in a 1 or 2-column grid each with a left color border, likely outcome as headline, best/worst outcome lines, risk and reversibility score badges, risk list, assumption callout. Strategic recommendation card and inaction card at the bottom. Archive saves past trees. Make it work in a single HTML file.What this does: Describe the decision you're facing, choose how many paths to map (2–4), set a time horizon, and note what matters most to you. The Decision Tree generates an actual visual SVG tree branching from a root decision node, with each path color-coded. Below the tree, each path gets a detailed card showing the most likely outcome, best and worst cases, risk level, reversibility score with an explanation of why, key specific risks, and the single assumption the path depends on most. A strategic recommendation synthesizes everything against what you said matters, and an inaction card shows what happens if you decide not to decide. Every tree saves to localStorage.
What this looks like:
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Decide who you date, by reading your private answers and matching you against millions of other people's profiles (Bumble's Bee).
❌ Still Can't: Build a working zero-day exploit on its own. Even with Claude Mythos, three skilled humans had to design the actual attack chain against Apple's M5 chip.
✅ AI Can Now: Drive your browser, click buttons, fill out forms, and extract data without sending any of it to a cloud server (Kimi WebBridge).
❌ Still Can't: Tell the difference between an employee actually working and an employee gaming a leaderboard with fake AI tasks. That's how Amazon ended up with "tokenmaxxing."
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A New York ad exec gets early access to augmented reality glasses and starts building an AI version of his best friend's girlfriend that he can interact with through the lenses. Shot in stark black and white by writer-director Benjamin Dickinson, it predicted the exact texture of the AI companion problem years before Replika or Character.AI existed.
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.
