Biased Dice, Papal Advice
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Monday, May 25, 2026
You know that moment when you're juggling twelve browser tabs, three Slack threads, and a calendar reminder you already missed?
Turns out, AI is having the same problem - except it's running twenty coding agents at once and still needs a human to decide what actually matters. This Monday brings us AI solving 56-year-old math puzzles for pocket change, the Pope weighing in on artificial intelligence, and proof that even the smartest models can't pick a random number to save their lives.
Today in AI:
- Your AI Can't Roll Dice Either - A researcher asked GPT-4.1 to pick a random number between 1 and 100 ten thousand times. The results cluster in suspiciously human patterns, suggesting the model inherited our biases rather than acting like a fair random generator. Microsoft Github
- Vibe Code Your Own Mac App Before Lunch - Glaze, a new Mac app from the Raycast team, lets you describe software in plain English and get a working local app in minutes. Unlike browser-based tools, these apps run offline and tap into your Mac's file system. Free with limits, or $20/month. Raycast
- Bug Bounties Just Got a Lot More Crowded - AI agents are now finding software vulnerabilities faster than human researchers, flooding bug bounty programs and reshaping cybersecurity economics. One researcher says he's submitted three times more bugs than last year. Attackers are using the same tools. Wired
- The Pope Has Thoughts on AI - Pope Leo XIV released a 42,300-word encyclical on Monday warning against equating AI with human intelligence, calling for regulation to prevent wealth concentration, and insisting humans - not AI - must control all weapons decisions. Anthropic co-founder Christopher Olah stood alongside him. Engadget
- Right Answer, Wrong Receipts - Peking University researchers found that leading AI models like GPT and Gemini routinely cite passages that don't actually support their answers. They call it "attribution hallucination," and it's a serious problem for law, medicine, and any field where you need to show your work. The Decoder
- AI Solves 56-Year-Old Math Problems for a Few Hundred Bucks - Google DeepMind's AlphaProof Nexus autonomously cracked nine open Erdos problems, including two that stumped mathematicians for decades. Each proof cost just a few hundred dollars in compute, though the overall success rate is a humble 2.5 percent. The Decoder
- Twenty Agents, One Brain - Developer Addy Osmani coined the concept of "orchestration tax": spinning up AI agents is cheap, but reviewing and merging their output still bottlenecks through your single human brain. His advice - architect your attention the way you'd architect a concurrent system. Addy Osmani
- Amazon's Bee Wearable Records Your Life, For Better or Worse - TechCrunch tested Amazon's Bee, an AI wrist device that records, transcribes, and summarizes your conversations all day. It shines in meeting-heavy workdays but raises real privacy questions about wearing an always-listening gadget. TechCrunch
Today's Takeaway:
Here's the thread connecting Monday's biggest stories: AI keeps getting more capable, but the bottleneck is shifting squarely onto humans. AlphaProof Nexus solves decades-old math problems for pocket change - but only lands 2.5 percent of the time, according to The Decoder. AI models nail the answer but cite phantom evidence, per Peking University researchers. Bug-hunting agents flood programs with findings, but someone still has to triage every report, as Wired details. The pattern is unmistakable: generating output is now trivially cheap, but verifying, curating, and deciding what to do with it is not.
Addy Osmani's "orchestration tax" concept frames this perfectly. Think of it like hiring twenty interns who work at superhuman speed but can't tell the difference between brilliant and broken. You still have to review every deliverable. That means the most valuable skill in an AI-powered workplace isn't prompting - it's judgment. First order: AI produces more. Second order: humans spend more time reviewing. Third order: the people who build strong verification habits will pull dramatically ahead of those who just spin up more agents. If your Monday goal is one thing, make it this: get better at checking AI's homework, not just assigning it.
📋 Try This
Research shows humans tend to pick the same 'random' numbers - and AI has its own predictable patterns too. These prompts help you explore how AI thinks about randomness and prediction, inspired by today's story on GPT number-guessing behavior.
For Business Owners:
I want to understand predictable patterns in decision-making. Ask me to pick a random [NUMBER, COLOR, WORD, OR CHOICE] from a list, then explain what most people typically choose and why. Then tell me how a [BUSINESS TYPE] like mine could use knowledge of these predictable patterns to make better decisions about [MARKETING, PRICING, OR PRODUCT DESIGN].
For Personal Use:
Ask me to pick a random number between 1 and 100. After I answer, tell me how common my choice is and why humans tend to pick it. Then help me understand one area of my daily life - like [SHOPPING, MEAL PLANNING, OR SCHEDULING] - where my choices might be more predictable than I think, and suggest one small way to break the pattern.
💡 Copy either prompt, swap the brackets with your own details, and paste it into ChatGPT or any AI chat tool.
The Bottom Line
The Pattern: AI output is exploding - in code, in math proofs, in bug reports, in conversation transcripts. But every piece of output still needs a human to decide if it's correct, relevant, and worth acting on. The bottleneck has officially moved from creation to curation.
Why It Matters: If you're a business owner, a freelancer, or just someone trying to stay competitive, the temptation is to run more AI tools faster. But as today's stories show, speed without verification creates attribution hallucinations, flooded inboxes, and orchestration fatigue. The real advantage goes to whoever builds the best judgment muscle.
Your Move: This week, pick one AI output you normally accept at face value - a summary, a citation, a code suggestion - and actually verify it. You might be surprised how often the answer is right but the reasoning is wrong. That gap is where your value lives.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
✨ How AI-Fluent Are You? - Test your AI fluency with our interactive quiz. See how you stack up and discover what to learn next. → Take the Quiz
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team