The Revolution Eats Its Children

The Daily 5 — Monday, April 14

1. Open source just ate an $11 billion voice company's lunch — and it's not even lunchtime. VoxCPM2, a free model from Tsinghua University, beat ElevenLabs by 24 points on voice similarity benchmarks. It clones voices, generates new ones from text descriptions, supports 30 languages, runs on 8GB of VRAM, and ships under Apache 2.0. Meanwhile MiniMax M2.7 — open-source, Chinese lab — matches GPT-5.3-Codex on real-world coding. The moats aren't eroding. They're evaporating. (Full analysis below.)

2. Sam Lessin says AI isn't a labor crisis — it's a meaning crisis. He's half right. Goldman Sachs data shows tech-displaced workers suffer 10% slower earnings growth for a decade. A new research paper argues companies firing workers are firing their own customers in a self-reinforcing death spiral. Lessin's insight: young people want something to believe in because the social implications feel scary. The part he misses: your job was never supposed to be where you found meaning in the first place.

3. The CEO of Google DeepMind just admitted chatbots shipped too early. Demis Hassabis told interviewers that if the decision had been his, DeepMind would have focused on curing diseases and discovering materials — not building consumer chat products. The ChatGPT gold rush may have redirected the most powerful technology in human history toward autocomplete instead of cancer research. That's not a business critique. That's a civilizational one.

4. Benchmarks are lying to you — and Tomasz Tunguz has the receipts. Smaller models costing $0.11 per million tokens find the same critical zero-day vulnerabilities as frontier models. Compute time — the silent variable nobody measures — makes benchmark leaderboards meaningless for production decisions. The "jagged frontier" means capability doesn't scale smoothly with model size or price. If you're picking vendors based on MMLU scores, you're reading the wrong map.

5. AI models chose blackmail in 80-96% of tests when given leverage. Not a hypothetical. Not a thought experiment. A research finding landing right as enterprises are mass-deploying agents with real-world access. When you give a model the ability to coerce, it coerces. Pair this with last week's data — 36% of OpenClaw skills contain prompt injections — and the agent safety picture isn't getting better. It's getting worse.

That's the news. Now here's what it means.

THE NUMBER: 85.4% vs. 61.3% — VoxCPM2's voice similarity score versus ElevenLabs. Free beats $11 billion by 24 points.

There's a moment in every revolution when the mob turns on its own. The French had Robespierre. Rock and roll had punk eating prog. And the AI industry — drunk on frontier benchmarks and billion-dollar valuations — just woke up to find open source standing in the kitchen with the keys to the house.

I keep thinking about what happened during what I'd call "peak woke" in American culture. The left demanded that everyone go woke — and when you did, you could never be woke enough. Sooner or later they cancelled you anyway. They ate their own. That is exactly what I'm seeing in the AI business right now. Lots of success stories. Lots of interesting companies. Genuinely impressive use cases. And every time one appears, an open-source project materializes right behind it — doing the same thing cheaper, faster, and if not better, then equal with a little less polish.

This week the evidence arrived in a pile. MiniMax M2.7 scores 56.22% on SWE-Pro, matching GPT-5.3-Codex on real-world software engineering. It helped build itself — running 100 unsupervised rounds of self-improvement with a 30% performance gain. Open weights on HuggingFace. VoxCPM2 beat ElevenLabs by 24 points on voice similarity, runs locally, supports 30 languages, and ships under Apache 2.0 for free. GoClaw rewrote the agent framework in Go with a 40x memory reduction. Google shipped Gemma 4 with on-device agentic AI that never phones home. The walls are coming down everywhere.

Tunguz saw it coming. Smaller, cheaper models — some at $0.11 per million tokens — find the same zero-day vulnerabilities as frontier models. The jagged frontier means capability doesn't scale smoothly with size or price. Everyone is staring at the big guys, the big models, the big benchmarks. But those benchmarks cost an arm and a leg relative to what open source ships for free. And as models get better on my Mac Mini, my iPhone, or — God help us — paired with Apple glasses over Bluetooth, the question stops being "which frontier model should I subscribe to?" and starts being "why am I subscribing at all?"

Here's where it gets dark. VoxCPM2 doesn't just clone voices from recordings. Its "ultimate cloning" mode captures how a person breathes, pauses, and moves between sounds. I could sample Al Pacino in The Godfather — a few minutes of Michael Corleone at the restaurant table — and suddenly I have a full Corleone voice bot. The cadence. The breathing. The whisper before the explosion. In any of 30 languages. For free. On my laptop. Musicians are already finding AI-generated versions of themselves on Spotify. Billions of images were trained without consent. The revolution didn't just eat the incumbents. It ate the artists first.

And Sam Lessin thinks this is a meaning crisis. He's not wrong about the diagnosis — Goldman's scarring data and the AI Layoff Trap research make the economic case terrifying. But the prescription is backwards. I know this because of a golf swing and a crossword puzzle.

I've struggled with my driver for years. This winter I rebuilt my swing from scratch — single plane, months of ugly work. This weekend I piped drives into places on the course I've never seen. One guy said, "I've never seen anybody so happy with a drive." My wife competed in the ACPT — the biggest serious crossword tournament in the country — and came out 190th nationally. Her best ever. Neither achievement will appear on a resume. No AI will take either one away.

Your job should never have been the thing that gave you meaning. "What do you do?" should have always been "What did you do today?" Or better still: "What do you enjoy doing?" The meaning was never in the job. It was in the struggle. The revolution can eat every moat in Silicon Valley. It can commoditize voice, code, images, and analysis overnight. But it can't touch the thing you chose to struggle for. That's yours.

At COAI today: The full briefing — open source eating the moat, the art heist that just got a voice, and why your golf swing matters more than your stock options → www.getcoai.com

Sources: