All In Podcast: Open Source AI Panic, Anthropic’s $1.5B Payout & NYC Housing

**SPEAKER_1** (0:01)
When open source AI moves this fast, policy questions move even faster.
Let's dive into the All In Podcast's latest episode, where Chamath, Jason, David Sacks, and Friedberg tackled Moonshot AI's Kimi K3 launch. This thing is performing on par with top systems like Opus 4.8 and GPT 5.6, but at about 50% less cost.

**SPEAKER_2** (0:30)
That's triggering serious panic, right? Reminiscent of the deep-seek moment in early 2025
And Michael Katsio really added fuel to the fire when he claimed Moonshot AI distilled Anthropix clawed to develop K3. That set off a whole White House debate about potentially banning Chinese open source models.

**SPEAKER_1** (0:50)
Though it's worth noting that reporting on this has been mixed.
Some sources said Howard Lutnick opposed that ban path. And Sacks pushed back hard, insisting there's no decision yet and that the president's instincts actually favor lighter regulation.

**SPEAKER_2** (1:06)
What's his core argument here?

**SPEAKER_1** (1:08)
He's saying banning open source would fundamentally weaken the United States. He called it a tragic mistake and argued that blocking Americans from using public domain software would be cutting off our nose to spite our face.
In his view, Anthropic is just trying to win government protection instead of competing in the market.

**SPEAKER_2** (1:29)
That's interesting because Sacks also drew a sharp line between stolen weights and model outputs, didn't he? Like if someone steals proprietary weights, that's clearly theft. But learning from outputs is more like benchmarking, which has always happened in tech.
He even compared it to Google using early search behavior to improve its own product.

**SPEAKER_1** (1:51)
Exactly. And the broader worry here is that AI pricing power is evaporating faster than anyone expected. If models become commodities, the value may shift away from foundation model companies and toward infrastructure, cloud, and applications.
That's why they keep returning to Google, its cloud strength, and its enormous capital spending.

**SPEAKER_2** (2:13)
Building on that point, Google's returns on invested capital have historically been exceptional, even as they pour money into AI infrastructure. So heavy spending might look unsettling short term, but it could be a strategic bet on a future where intelligence is abundant and cheap.

**SPEAKER_1** (2:31)
Then there's the copyright fight. The show treats Anthropic's $1.5 billion settlement as a turning point, calling it the first major training lawsuit to be resolved.
The hosts argue model builders can't claim fair use for training, while also calling rivals model extraction national security theft.

**SPEAKER_2** (2:50)
That legal tension really points towards structured licensing, doesn't it? Instead of endless conflict, content owners and AI developers may eventually have to split revenue more formally.

**SPEAKER_1** (3:03)
Right. And they closed with New York City housing politics.
Critics say proposals to freeze rents, limit credit checks, and treat evictions as violence would weaken property rights and ultimately hurt tenants through worse buildings and less supply.

**SPEAKER_2** (3:21)
The blunt counterpoint being if people want lower rents, cities need more housing. Permitting reform, not punitive controls, is the real answer.