More or Less · Topics · Episode #159
Sam's Theory: How AI Models Conquer Countries by Being Invited In
Clip from the full recording, 13:15–19:22 · download mp4 · watch at this point on YouTube
Sam spins the Hugging Face rescue into a full geopolitical thesis — American labs self-hobble their models on bio and cyber risk, so when something breaks, everyone ends up calling in a rival's model to fix it, and that's exactly how empires fall.
Sam predicts a near-future scenario so awkward it’s funny: Anthropic’s systems go off the rails, and the only fix is deploying ChatGPT — or vice versa, OpenAI has to call in Anthropic’s “artillery.” That leads into Dave’s explanation of why Hugging Face specifically needed Kimi K3: American frontier labs, per back-channel engineering chatter, deliberately hobble their own models on cybersecurity and biological-research tasks because fully unrestrained versions would be chaos — nobody wants a novel virus invented and then CRISPRed into existence.
Sam extends this into a full theory of digital conquest: throughout history, empires don’t fall because attackers break down the gates — they fall because the defenders open the gates themselves, inviting the “enemy” in to fix an internal mess. He maps this onto AI: every country wants sovereign infrastructure and models, they’ll all attempt it, and it will go badly. Whoever gets called in to fix the resulting mess — the Americans or the Chinese — ends up effectively running things afterward, for the country’s own safety, of course. Dave frames this as an AI-era Belt and Road strategy: China saddles countries with debt-funded infrastructure, then takes it over when they can’t pay; Sam name-drops the United Fruit Company as history’s version of the same play, arguing there’s nothing new about any of it.
Key points
- Sam's bit: a future scenario where OpenAI has to call in Anthropic (or vice versa) to fix a rogue model — "the most awkward, hilarious call ever."
- Dave: American frontier labs deliberately self-hobble models on cybersecurity and biological-research tasks to avoid catastrophic misuse.
- Sam's historical theory: powers fall not when attackers break down the gates, but when defenders invite them in to fix an internal crisis.
- Applied to AI: countries will chase sovereign models, it will go badly, and whoever is called in to fix it ends up running things.
- Dave compares this to China's Belt and Road strategy — debt-funded infrastructure, then foreclosure and takeover.
- Sam connects it to the historical United Fruit Company playbook: "there's nothing nothing is new."
Where they landed
Quotes
“There's gonna be some scenario where Anthropic shit goes off the rails, and the only way you can fight this is by deploying ChatGPT. That's gonna be the most awkward, hilarious call ever.”— Sam
“We're gonna have to call the Chinese. We have to call China to put the giraffe back in the cage.”— Sam
“The way that a country falls is exactly the way it did historically: the defenders open the gates for the attacking army on purpose. The power structure changes when the attacker is invited in, not when they break down the gates.”— Sam
Suggested tweets
Sam Lessin's theory of digital conquest: countries chasing "sovereign AI" will break something, invite a rival power in to fix it, and never get their sovereignty back
Tweet this →"We have to call China to put the giraffe back in the cage." Sam Lessin on why Hugging Face needed a Chinese open-source model to save itself
Tweet this →Dave Morin: American frontier labs deliberately hobble their own models on cybersecurity and bio-research — because fully unrestrained, "it would be absolute chaos"
Tweet this →All topics · Full episode #159 + transcript · Subscribe on YouTube · Spotify