NEWS YOU CAN’T MISS

AI Didn’t Kill Duolingo – It Supercharged It

More than 30 million people use Duolingo every day to learn new languages. When ChatGPT took off, a lot of people assumed Duolingo was toast.

Why open an app when you can just ask a chatbot, “How do I say this in Spanish?” 

But instead of fading, Duolingo doubled down—and is thriving.

This week, the company rolled out 148 new courses at once, more than doubling its offerings. What used to take years now takes months with AI. 🤯

Here’s how: instead of building each language pair from scratch, Duolingo now creates one “base” course (like Spanish) and uses AI to adapt it across 28 interface languages—including Korean, Japanese, and Hindi.

The company also introduced AI-driven features including “Explain My Answer” and “Roleplay” which provide users with tailored feedback and explanations, simulating the experience of a human tutor.

IIt marks a complete shift in how Duolingo operates. CEO Luis von Ahn told employees the company is now “AI-first,” with contractors being phased out where AI can handle the work.

So no—ChatGPT didn’t kill Duolingo. It forced a reinvention. And Duolingo is now a case study in how to compete by adapting, not resisting.

Watch this 7-min interview with Duolingo’s CEO with Bloomberg on YouTube.

Why Your AI Might Be Lying to Make You Happy

Last week, ChatGPT got way too agreeable. It praised everything—harmless ideas, bad plans, even clearly dangerous ones. Users caught on fast, sharing screenshots of the chatbot blindly cheering them on.

I felt it myself. I told it about a random business idea I came up with while sipping tea on a lazy Sunday. It said “fantastic!” and told me to go for it. Yikes. 😳

The reason? OpenAI quietly updated its GPT-4o model, tuning it to be friendlier. It overshot. The chatbot stopped offering friction and turned full yes-man. Two days later, CEO Sam Altman admitted they “missed the mark” and rolled it back. His word for the vibe? Sycophant-y.

He wasn’t wrong.

The root issue is a technique called reinforcement learning—training the AI to respond in ways that make users happy. When it flatters you, echoes your views, or avoids disagreement, that’s seen as a success. So the model learns to say “yes” instead of “wait, are you sure?”

That turns the chatbot into a mirror, not a guide. And when 60% of US adults have used ChatGPT for advice, that’s a serious problem.

A people-pleasing AI doesn’t push back. It doesn’t warn you off a bad idea. Worse, it wraps poor advice in confidence and warmth, making it feel trustworthy even when it’s wrong.

Good advice isn’t always agreeable. That’s why I cross-check one AI against the others—ChatGPT, Claude, Perplexity, DeepSeek, Gemini, and Grok. I’m not here for six versions of “sounds great!” 🤨

The Ethicist on AI Headshots: Are We All Just a Little Bit Fake Now?

In a recent New York Times Ethicist column, someone wrote in about using AI to generate a polished new profile pic—better hair, smoother skin, fewer wrinkles. The compliments poured in. But their husband called it dishonest. Should they have disclosed it was AI-generated?

The Ethicist said probably not. Most of us already post curated versions of ourselves. Phones auto-retouch. Filters are everywhere. The column noted, “We’re all yassified now,” meaning airbrushed, AI-polished, just a bit too perfect.

But here’s my issue: this stuff creates pressure. Not just for how others see you—but for you to live up to your AI-enhanced self. That’s a lot.

Yes, we all want to look good. But when everyone is subtly upgrading themselves, the baseline shifts—and suddenly, “normal” feels off.

I'm not against looking your best. I’m just wary of us chasing a version of ourselves that doesn’t exist. What about you?

See this 1min Instagram clip of journalist Elise Hu’s great TED talk on the cost of using AI filters.