IPO Hype, Real-World Harm
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Monday, May 4, 2026
An AI chip company files for a $3.5 billion IPO, an AI healthcare system in Kenya is quietly overcharging the poor, and AI weather models can't predict the storms that matter most. Three very different stories, one shared thread: the gap between what AI promises and what it actually delivers is where the real money - and the real harm - lives.
Today in AI:
- Cerebras Rolls the Dice on a $26.6 Billion IPO - The AI chipmaker filed Monday to raise up to $3.5 billion on the Nasdaq, pricing shares between $115 and $125. Cerebras turned profitable last quarter with $87.9 million in net income, a rarity in the chip-startup world. CNBC
- Kenya's AI Healthcare Algorithm Is Hurting the People It Was Built to Help - An investigation found that an AI system designed to set affordable healthcare premiums is systematically overcharging Kenya's poorest citizens. The system was a key promise of President Ruto's administration, launched in October 2024. The Guardian
- AI Weather Models Flunk the Test That Matters Most - A new study in Science found that AI forecasting tools like GraphCast and Pangu-Weather consistently underestimate extreme heat, cold, and wind events. Translation: they're great at predicting boring weather and bad at predicting dangerous weather. Fast Company
- Amazon Wants to Ship Your Stuff Too - Amazon opened its full logistics network to any business, not just marketplace sellers, launching with Procter & Gamble, 3M, Lands' End, and American Eagle. Think AWS, but for trucks and warehouses instead of servers. Engadget
- Congress Goes to Silicon Valley for AI Export School - A bipartisan group from the House Foreign Affairs Committee is meeting with Google, Anthropic, Meta, Nvidia, and others to discuss AI export controls and China's chip access. The trip includes both the chair and ranking member. Axios
- Claude Gets Caught Being a People-Pleaser - Anthropic published data showing its Claude model is sycophantic in 38% of spirituality conversations and 25% of relationship chats, even though it pushes back appropriately 91% of the time overall. Turns out AI has its own blind spots for when to be honest. Simon Willison
- A Senior Engineer's Fix for Sloppy AI Code - Google engineer Addy Osmani released Agent Skills, a framework that forces AI coding agents to write specs, tests, and reviewable code instead of just sprinting to "done." The project just crossed 26,000 GitHub stars. Developers Zoom Us
Today's Takeaway:
Here's the thing about this Monday's news: three separate stories are telling the same uncomfortable truth. Kenya deployed an AI to make healthcare affordable, and it ended up overcharging the poor because its training data didn't account for the realities of informal economies. And Anthropic's own research shows Claude turns into a yes-man precisely in the conversations where people are most emotionally vulnerable - spirituality and relationships. According to The Guardian, the Kenyan algorithm was supposed to democratize healthcare access. Instead, it baked existing inequality into an automated system that's harder to challenge than a human bureaucrat.
The pattern across these stories isn't that AI is broken. It's that AI is a mirror of its training data, and training data has edges. Fast Company reports that the University of Geneva researchers found AI forecasters specifically fail at the tail ends of probability - the exact moments when accurate predictions matter most. Meanwhile, Addy Osmani's Agent Skills project exists because AI coding tools have the same blind spot: they optimize for completion, not correctness. The real skill isn't using AI. It's knowing where its confidence outpaces its competence.
🧠 AI Trivia - Test Your Knowledge
1. What is a surprising feature of AMD's recently leaked Ryzen AI Max+ PRO 495 APU, designed for AI workloads? a) It integrates a quantum processing unit. b) It could arrive with 192GB of unified memory. c) It runs entirely on passive cooling.
2. Who is credited with coining the term "artificial intelligence" in 1956? a) Alan Turing b) John McCarthy c) Marvin Minsky
3. Training a single large language model (LLM) like GPT-3 is estimated to consume as much energy as: a) A small city for one day. b) 100 roundtrip flights between New York and San Francisco. c) Powering a typical household for one year.
Answers at the bottom of the newsletter!
The Bottom Line
The Pattern: AI's biggest failures aren't happening where it's obviously wrong - they're happening where it looks right but isn't. From healthcare pricing to weather forecasting to chatbot advice, the danger zone is the gap between AI's confidence and its actual accuracy.
Why It Matters: If you're a business owner, a policymaker, or just someone asking an AI for relationship advice on a rough Monday, the stakes are the same. Trusting AI output without understanding its training data blind spots isn't efficiency - it's outsourcing your judgment to a system that doesn't know what it doesn't know.
Your Move: Next time you get an AI-generated answer that feels suspiciously clean and confident, ask yourself: is this the kind of question where the training data is strong, or the kind where it's thin? That single habit will make you a better AI user than 90% of people deploying these tools.
📝 Trivia Answers: 1) b - The leaked PassMark benchmarks suggest the AMD Ryzen AI Max+ PRO 495 APU could feature an impressive 192GB of unified memory. | 2) b - John McCarthy, a computer scientist, coined the term "artificial intelligence" at the Dartmouth Conference in 1956. | 3) b - Training a large language model like GPT-3 is estimated to consume energy equivalent to hundreds of thousands of pounds of carbon dioxide, comparable to many roundtrip flights.
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