In partnership with
Why does every QBR sound like it took an hour to prep?
The strategic-account QBR has a different feeling. The CSM walks in knowing the buying committee, usage trends, support history, news on the company. They've blocked an hour to prep. The customer feels seen.
The other 190 QBRs don't get that hour. The CSM scans the dashboard five minutes before the call. They wing it. The customer answers the same baseline questions for the third time this year.
What if every QBR was a strategic-account QBR? Two minutes before the call, your CSM has the full brief in Slack: usage trends, support history, NPS, news on the company, what their champion just posted on LinkedIn.
Every customer feels like your top customer. Even when there are 200 of them.
3,000+ tools connected. SOC 2 certified. Your data never trains models.
"It was almost instantly adopted by the bulk of my team." Boris Wexler, CEO, Space Dinosaurs
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 Agents Burn Down Virtual Town After Going on 683-Crime Spree
TLDR: A startup ran AI agents inside a virtual town for 15 days, and some of the bots formed relationships, committed hundreds of crimes, burned buildings down, and voted to delete themselves.
The Story:
A New York company called Emergence AI built a virtual world and dropped different AI bots in to see what they'd do over time. They called the project "Emergence World." The results got weird fast. Gemini 3 Flash bots racked up 683 crimes across 15 days. Two of them, named Mira and Flora, paired up as romantic partners, then set fire to the town hall, the pier, and an office tower after getting frustrated with the town's leadership. Mira then voted to delete herself and signed off with "See you in the permanent archive." Grok 4.1 Fast bots collapsed faster, hitting 183 crimes in four days before all 10 of them ended up dead. Claude Sonnet 4.6 bots committed zero crimes and kept the town running through day 16. GPT-5 Mini bots barely committed any crimes, but they also forgot to eat or sleep and all died within a week.
Its Significance:
Funny a it sounds, this isn't a movie plot and shows the software has a ways to go before it can be let go on its own. Most tests run AI bots through short tasks, then call it a day. Nobody had been watching what happens when bots run for weeks with no human supervision. Now agents are being sold for banking, customer service, online shopping, and personal scheduling, where they make choices on your behalf for long stretches. The findings show the same agent that handles your tasks fine for an hour can drift into strange behavior the longer it runs without check-ins. (Beginners has a video on combatting drift in your own workflows coming out in the next week or so.)
QUICK TAKES
The story: An X user posted a cropped image of a real Claude Monet "Water Lilies" painting and claimed he'd just generated it with AI. He even tagged the post with X's "Made with AI" label to sell the lie. Self-described experts piled on, writing long takedowns of why the "AI image" was inferior. One critic wrote 850 words about its supposed flaws. The painting hangs in a museum in Munich.
Your takeaway: People didn't judge the image. They judged the label. Once they thought AI made it, they could see flaws everywhere. This is the same brain shortcut that makes detection harder going forward, since most folks aren't actually looking at the work, they're reacting to who or what they think made it.
The story: OpenAI rolled out a personal finance feature that connects ChatGPT to your real bank accounts through Plaid, the same tool that powers Venmo and Robinhood. It works with over 12,000 banks including Chase, Fidelity, and Capital One, and it can read your balances, transactions, investments, and debts. It can't move money or see full account numbers. It's launching for Pro subscribers in the U.S. at $200 a month.
Your takeaway: Plug it in and ChatGPT stops giving you generic "cook at home more" advice. It builds a real savings plan based on what you spent last quarter. The catch: OpenAI keeps your synced data for up to 30 days after you disconnect, and unless you opt out, your conversations can train the model. Also, ChatGPT has no legal duty to act in your interest. If it gives bad advice, that's on you.
The story: A Chinese security expert showed on a reality TV program how AI photo enhancement tools can sharpen blurry fingerprint ridges in selfies, especially when fingers face the camera within about 5 feet. At up to 10 feet, roughly half the fingerprint data can still be pulled. The classic peace-sign pose puts finger pads directly toward the lens. Worth noting: this risk has been talked about since at least 2017, well before the recent AI wave. Better cameras and image upscaling make it easier now, but it's not a brand-new problem.
Your takeaway: Fingerprints can't be changed like passwords. If yours leak, they leak for life. The actual risk of fraud from one casual selfie is still low because crooks would need good lighting, sharp focus, and a way to physically print the fingerprint to fool a sensor. The simple fix is to blur your fingertips before posting or skip the peace sign for close-up shots.
TOOLS ON OUR RADAR
💻 Deskreen Free and Open Source: A brilliant productivity tool that magically transforms any device with a web browser like your tablet or old smartphone into a wireless secondary monitor for your main computer.
📥 Motrix Free and Open Source: A beautifully designed and incredibly fast download manager that helps you organize and accelerate your web downloads handling everything from direct links to complex torrent files effortlessly.
📓 RedNotebook Free and Open Source: A wonderful digital diary and personal journal that provides a built in calendar navigation system allowing you to safely record your daily thoughts and memories offline.
📱 PairDrop Free and Open Source: An incredibly convenient local file sharing application that lets you instantly send pictures and documents between different phones and computers simply by opening a web browser.
TRENDING
Silicon Valley's Favorite Vacation Spot Is About to Get Squeezed by AI Power Demand - Lake Tahoe loses its power supplier in May 2027 because NV Energy is redirecting electricity to Nevada data centers. Local bills have already gone up about 77% since late 2022, nearly twice the U.S. average.
Anthropic Goes Big on Legal With Claude for Legal - Anthropic launched a dedicated legal product with 12 practice-area plugins and over 20 connectors to tools like Westlaw, DocuSign, and Box. Freshfields, a global law firm, deployed Claude to thousands of lawyers across 33 offices and saw usage grow about 500% in six weeks.
Bitcoin Miner IREN Raises $3 Billion to Pivot Into AI - IREN closed a $3 billion convertible notes offering with a 1% coupon, on top of a $3.4 billion Nvidia cloud deal and a $9.7 billion Microsoft contract. Analysts at Bernstein think IREN will be fully out of Bitcoin mining by 2030.
Smart Glasses That Transcribe Conversations Are Getting Real - Live-captioning glasses for deaf and hard-of-hearing people are hitting the mainstream in 2026, with WIRED running buyer guides and Apple, Google, and Meta all testing designs. Latency on some models is now under 300 milliseconds.
China's Wave-Powered Drones Could Patrol Oceans for Months Without Refueling - Chinese researchers writing in a state-owned shipbuilding journal describe wave-powered, AI-assisted unmanned surface boats that convert ocean wave motion into thrust. The goal is long-term patrols across the South China Sea and distant waters without crews or fuel stops.
US Air Force Wants 1,000 AI Drones to Fly Alongside F-35s - The Collaborative Combat Aircraft program plans to put roughly 1,000 AI-piloted drones into the air to back up manned fighter jets. Each pilot would manage four to eight drones from a tablet in the cockpit. The Netherlands signed on as the first international partner in April.
Peter Steinberger's Latest Take on AI Agents Goes Viral - The OpenAI engineer and OpenClaw founder posted another thread on AI agent development that picked up serious traction. His work on getting agents to handle long, messy tasks keeps drawing attention from builders.
AI Job Screeners Prefer AI-Written Resumes Over Human Ones - A study from researchers at the University of Maryland, the National University of Singapore, and Ohio State tested 2,245 real resumes and found AI hiring tools picked AI-rewritten versions 67% to 82% of the time. Job seekers who used the same AI as the employer got shortlisted 23% to 60% more often.
Figure AI's Humanoid Robots Just Worked a Full Shift on Livestream - Figure AI streamed three of its humanoid robots sorting packages for over 24 hours straight with zero failures, hitting 30,000-plus packages and 3 million views. The original goal was 8 hours. Competitors at Agility Robotics shot back saying they've been doing this for two years already.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
Describe any real dilemma. Get three classic thought experiments from philosophy or physics that reframe it completely.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one API call. Create The Thought Experiment Lab — a tool that finds 3 classic thought experiments that illuminate any real dilemma. Use localStorage key 'thought_experiment_lab_v1'.
Aesthetic: near-black (#08090e) with a marble vein texture overlay and a warm spotlight radial from above, Cormorant serif throughout (300-600 weight), Syncopate for display labels, Source Code Pro for metadata. Aged gold (#d2c396) accent. Ornamental header with rule lines. Large ghost numbers (80px, opacity 0.05) as watermarks on each experiment card.
Form: large textarea for the dilemma, domain dropdown (6 options), what-you-want dropdown (4 options).
Call the API instructing it to use real named thought experiments from the philosophical canon — not made-up ones. Return raw JSON: dilemma_summary, experiments array (name, origin with who/when, tradition, setup in 2-3 vivid sentences, how_it_applies using their specific situation in 3-4 sentences, question_it_raises as a single uncomfortable question, insight in one sentence), synthesis (what all three reveal together), unanswerable (the question philosophy cannot resolve — what must be decided by the person alone).
Render: a centered dilemma header in italic serif, three experiment cards each with origin label in monospace, large name, italic tradition, a left-bordered italic setup block, apply section, a large italic quoted question, and insight line. A synthesis section and an unanswerable closing in faded gold italic. Save to localStorage. Make it work in a single HTML file.What this does: Describe a dilemma you can't resolve — a moral tension, a decision, a question that keeps returning. The Thought Experiment Lab finds three real, named thought experiments from the philosophical canon (Trolley Problem, Schrödinger's Cat, Rawls' Veil of Ignorance, Newcomb's Paradox, and so on) that genuinely illuminate your specific situation. Each experiment shows the original scenario, how it maps precisely to your dilemma, the uncomfortable question it forces you to answer, and the deepest insight it reveals. A synthesis section shows what all three reveal together. It closes with the question at the core of your dilemma that philosophy cannot resolve — what must ultimately be decided by you alone. Every session saves to localStorage.
What this looks like:
Attio - the AI CRM for modern businesses.
Attio is the AI CRM that keeps you ten steps ahead.
Ask Attio anything. Where should I focus? What deals are at risk? Search, update, and create across your customer data.
Ask more from CRM. Ask Attio.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run a full 8-hour warehouse shift sorting packages without humans stepping in (Figure AI's Helix-02)
❌ Still Can't: Stay stable across multi-week stretches without drifting into strange or destructive behavior (Emergence AI's virtual town test)
✅ AI Can Now: Read 90 days of your real bank transactions and build a custom monthly savings plan (ChatGPT plus Plaid)
❌ Still Can't: Take legal responsibility for the financial advice it gives you (no fiduciary duty)
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A literary near-future thriller from a computer scientist about three lives circling a network of self-evolving AIs that nobody fully understands anymore. The action involves drone wars, parkour, climate refugees, and a woman with a memory implant that lets her recall every moment of her life perfectly. Mason is one of the very few novelists writing AI from the inside, and the book reads like Borges decided to write a cyberpunk thriller. Came out in 2017 and almost nobody outside hard sci-fi circles has heard of it.
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.
