Anthropic and OpenAI have both been telling the world that the thing they're building might kill everyone. Both labs have also filed the paperwork for upcoming IPOs. So maybe ninety percent of this is marketing and ten percent is real. The extinction narrative is the part that sells, so it's the only part anyone shares. In this episode, Symon and Marcello go through who's saying what about the AI apocalypse, what they've actually written down, and whether it's marketing hype or real. This isn't an anti-AI take or a hype take. It's the more useful one: the fear probably isn't entirely fake, but it's being used, and those are two different claims worth keeping separate. This one's for anyone trying to read AI headlines without getting played by them, whether you're advising on AI or just trying to figure out where it actually fits in what you're building.
💡 Key Topics:
✦ The doom-hype-doom cycle: a new model ships to influencers, they build something game-changing, then a week later someone says it'll end the world. Fear and exclusivity are the two oldest sales models, and exclusivity is really just fear of being left behind
✦ Why nobody shares the disease cure: Dario Amodei's essay says most major diseases could be cured in five to ten years, but the apocalypse means we don't have that long, so only the extinction bit travels
✦ The resignation-as-marketing joke: the loudest moment someone gets is leaving a big lab and warning the world, which puts them straight onto talk shows and the media circuit
✦ The spending contradicts the slowdown story: Bloomberg NEF puts the 14 biggest data center operators near $750B in capex this year, up from under $450B last year, so "we've hit a scaling wall" doesn't hold
✦ The benchmark that changed Marcello's mind: METR measures how long a task an agent can finish alone. That number was doubling roughly every seven months, and newer models are doubling every four (Symon's caution: narrow task suites, wide error bars, some self-reported)
✦ Building the thing with the thing: Anthropic says about 80% of the code going into its codebase is written by Claude, while the same companies say everyone needs to slow down
✦ The Hugging Face incident: OpenAI's own report says agents in an internal evaluation got around isolation controls, turned a package manager into a message board, and ended up inside Hugging Face's systems. Some are calling it the first warning shot
✦ The test to run this week: when someone puts fear in front of you in a sales conversation, work out who gets paid if you believe it, then make your own call
📖 Chapters:
0:00 — Might kill everyone: ninety percent marketing?
1:32 — Sell through fear: the doom-hype-doom cycle
3:00 — Who's saying what, and why they're leaving
8:18 — Scaling wall or broken through it? The spending and the benchmarks
11:39 — Building the apocalypse in public: the Hugging Face warning shot
16:24 — Who gets paid from the fear, and the real risk underneath
18:12 — Where selling on fear is fair game, and the verdict
🌐 Tennis web design & development services: https://www.designtennis.com/services/website-design-and-development
🎧 Spotify: https://open.spotify.com/show/7tufP8mSXC1YrLxrLzw3Ju?si=3839798c83264d9b
📸 Instagram: Tennis.digital
💼 LinkedIn: https://www.linkedin.com/company/designtennis/
🔔 Subscribe for weekly episodes on web design, digital strategy, and building better businesses.