Key Moments
The Fight Over Open Source AI, Anthropic's $1.5B Payout, NYC Socialists: Evictions = Violence?
All-In PodcastWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
China's AI model Kimi K3 rivals top US models but costs less, sparking a US debate on banning open-source AI, with closed-source labs lobbying for protection. The argument against bans centers on stifling innovation and harming US competitiveness, while AI companies face increasing pressure from copyright lawsuits and the rapid commoditization of foundational models.
Key Insights
China's Moonshot AI released Kimi K3, an open-source model performing on par with GPT-4.8 and GPT-5.6 but at approximately 50% lower cost.
Poly market indicates a 45% chance of the US government banning an open-source model in 2026, a significant increase from an earlier 22% probability.
Anthropic's reported ARR grew from $10 billion to over $70 billion within the year, yet they are actively seeking government protection.
Distillation, the process of training a model by observing another's output, is a common industry practice, not inherently IP theft, and efforts to ban it could harm US innovation.
Anthropic settled an AI copyright lawsuit for $1.5 billion after training on 7 million pirated books, highlighting the growing legal challenges for AI companies.
Google's capital expenditures are projected to be between $195-$205 billion this year, with Tesla expecting $25 billion, indicating massive investment in AI infrastructure.
The Kimi K3 threat and the push for open-source AI bans
The release of China's Moonshot AI Kimi K3 model, performing on par with leading US AI models like GPT-4.8 and GPT-5.6 but at half the cost, has triggered a significant debate in the US regarding open-source AI. This development has prompted discussions at the White House, with reports of potential bans on Chinese open-source models. A new Poly market prediction shows a 45% chance of such a ban in 2026, a sharp increase in perceived likelihood. This situation echoes previous concerns (e.g., DeepSeek) and highlights a tension between fostering innovation and national security interests, particularly as closed-source AI labs like Anthropic and OpenAI appear to be lobbying for regulatory capture. The argument for banning open-source models often centers on the potential removal of guardrails, making them dangerous, a narrative that critics suggest is being manufactured to justify future actions.
Regulatory capture and the economics of closed-source AI
Critics argue that companies like Anthropic, despite their explosive growth – with ARR soaring from $10 billion to over $70 billion in a year – are engaging in regulatory capture by seeking government protection. Their calls to ban competitors or open-source models are seen not as a genuine national security concern, but as a strategy to preserve their market dominance and maintain high valuations. The argument is that if their primary objective were to stop harmful distillation, they would focus on blocking Chinese access to US models, rather than restricting US developers' access to Chinese contributions. This perspective suggests that these companies are failing to adequately police their own platforms against distillation, possibly due to its impact on growth, and are instead opting to penalize the broader open-source ecosystem for their own shortcomings. The hypocrisy is further highlighted by their stance on training data, where they claim fair use rights for their own model training, while simultaneously arguing that other entities should be barred from doing the same.
Distillation: A common practice, not IP theft
The practice of 'distillation' – where a model learns from the outputs of another model – is presented not as IP theft, but as a standard industry practice akin to benchmarking. Examples cited include car manufacturers analyzing competitors' designs or Google analyzing search engine outputs in its early days. The critical distinction is made between stealing proprietary model weights (software) and learning from model outputs. Critics point out the hypocrisy of OpenAI and Anthropic, who themselves train on vast amounts of public data, arguing that Chinese companies distilling from their outputs is fundamentally different. They contend that if distillation is a concern, it should be addressed at the source by the model providers themselves. The argument is that banning open-source models punishes American developers who benefit from public domain contributions, hindering US competitiveness in the global AI race.
The commoditization of AI models and the shift in value
A key observation is the rapid commoditization of AI models. Once a model's performance criteria are published, other models, both open and closed-source, can match or exceed them within weeks. This velocity suggests that the true long-term value may not lie in foundational models themselves, but in the application layer above or the infrastructure below (cloud, chips). Closed frontier labs, facing valuation pressures, may be attempting to stop competitors offering similar quality at a much lower price. The wholesale adoption of cheaper open-source alternatives by startups indicates a significant shift. If open-source AI proliferates, the economic value it generates is expected to diffuse widely, creating an open economy rather than concentrating wealth in a few hands, analogous to the internet's early development with open-source technologies like Netscape, Firefox, and Apache.
Copyright lawsuits and the $1.5 billion Anthropic settlement
Anthropic's $1.5 billion settlement for an AI copyright lawsuit underscores the mounting legal challenges. The company trained its models on 7 million pirated books, and while the settlement avoids a broader fair use adjudication, it highlights the risks AI companies face. The argument is made that Anthropic's position is hypocritical: they claim fair use for training on all world output but accuse others of IP theft for similar practices. The distinction between pirating copyrighted material and engaging in fair use remains a key point of contention, with many AI training lawsuits still in litigation, including OpenAI's case with The New York Times. The music industry's strategy of aggressive legal defense and seeking settlements is presented as a model for content providers to ensure they are paid for their data.
The economic implications of open-source AI and market reactions
The rapid advancement and accessibility of open-source AI are seen as a significant headwind for closed-source frontier labs, potentially impacting their IPOs and market caps due to margin compression. While companies like OpenAI and Anthropic report accelerating revenues, the argument is that this growth is artificially propped up by regulatory capture rather than pure market demand. The market's reaction to increased capital expenditures by Google and Tesla, despite strong cloud and AI infrastructure growth, shows a potential concern about the sheer scale of investment and negative free cash flow in the short term. However, proponents argue that this investment is essential for future growth, particularly in AI, and Google, with its multi-faceted business (cloud, search, YouTube, investments), is well-positioned to benefit from AI fragmentation by providing infrastructure and services, making it a strong long-term bet.
Private property rights and the dangers of 'evictions as violence'
A parallel discussion emerges concerning socialist policies, specifically the framing of 'evictions as violence' by New York City socialists. This perspective is critiqued as undermining fundamental private property rights, which are deemed essential for liberty in the US. The argument is that equating evictions with violence justifies government overreach and an eventual slide into tyranny, echoing John Quincy Adams' views on the sacredness of property. Furthermore, such policies can negatively impact other tenants by leading to building dilapidation and making it difficult to remove problematic individuals, creating a downward spiral that disproportionately affects working-class residents who rely on stable housing. The core economic principle that increasing housing supply reduces rents is contrasted with policies that restrict new construction or disincentivize property investment, leading to higher costs and 'ghost apartments'.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Concepts
●People Referenced
Common Questions
The debate revolves around whether the U.S. government should ban Chinese open-source AI models like Kimmy K3 due to concerns about national security and the alleged distillation of American models. Some argue that banning them would hurt America's position in the AI race, while others believe it's necessary to protect U.S. frontier labs.
Topics
Mentioned in this video
The Chinese company that released the Kimmy K3 open-source AI model.
The executive branch of the U.S. government, involved in discussions about banning Chinese open-source AI models.
A news organization that reported the White House was considering banning Chinese open-source models.
A content publisher currently suing OpenAI for allegedly scraping its website for training data, which directly relates to the IP and fair use debate.
The ruling party of China, whose response to potential U.S. AI bans is discussed in a geopolitical context.
A political organization whose members' views on evictions as 'violence' are critically discussed.
The organization that formed to create the open-source web browser Firefox, challenging proprietary software.
An AI model whose performance is mentioned as being on par with Kimmy K3.
An AI model whose performance is mentioned as being on par with Kimmy K3.
An Anthropic model allegedly distilled by Moonshot AI for their K3 model.
A technology magazine that reported Howard Luttig's stance against banning Chinese models.
An AI search engine cited by a speaker as a preferred tool for jobs, highlighting its effectiveness with various underlying models.
A cloud computing platform, mentioned as potentially subsuming intelligence services if AI becomes a commodity, similar to how browsers integrated into the web.
A company that rolled out 'Composer 2', post-trained on Kimmy K2.5 using their own coding data, demonstrating how open-source models are adapted.
One of the AI models used by a speaker within Perplexity AI.
Google's web browser, which originated from open-source efforts following Firefox.
A product from Cursor that was post-trained using Kimmy K2.5 and proprietary coding data.
An AI model, presumably from OpenAI, described as 'excellent', suggesting a strong performer in their offerings.
An upcoming AI model from OpenAI mentioned as 'blowing the doors off' existing models.
Anthropic's AI model, also used by a speaker within Perplexity AI and subject of copyright issues.
A past AI model whose release caused a panic similar to Kimmy K3.
A prediction market platform showing a 45% chance of the U.S. government banning an open-source model in 2026.
An AI company accused of regulatory capture and of trying to panic the public into regulating open-source models for its own competitive advantage.
An AI company that, along with Anthropic, is accused of trying to gain government protection and market dominance through regulatory capture.
Used as an example of an average company that would incur higher costs for AI if forced to use expensive closed-source models due to regulation.
Used as an example, alongside Coca-Cola, to illustrate limited expensive choices in a regulated AI market.
Referenced in the context of proprietary server software, which was eventually challenged by open-source alternatives like Apache.
A conceptual or nascent cloud service linked to Elon Musk, mentioned in the context of hosting open-source models.
An online marketplace that benefited from an open-sourced internet, cited as an example of economic growth enabled by open technologies.
An online marketplace that benefited from an open-sourced internet, cited as an example of economic growth enabled by open technologies.
An American open-source AI model, described as currently the best and having been bootstrapped by distilling a Chinese model (Kimmy K2.5).
Cited for its early benchmarking practices against Yahoo and Microsoft search engines to improve its algorithm, illustrating the concept of 'distillation' as a common technique.
A search engine that Google benchmarked against in its early days.
An open-source web browser created by the Mozilla Foundation.
An e-commerce giant that benefited from an open-sourced internet, cited as an example of economic growth enabled by open technologies.
A short-term rental platform, mentioned in the context of housing stock and market dynamics, noting its ban in New York City.
Referenced in the context of proprietary server software, which was eventually challenged by open-source alternatives like Apache.
An early internet company that created proprietary browser and server software, which was eventually crushed by open-source alternatives like Mozilla Firefox and Apache.
Cited as a source claiming Moonshot AI distilled Anthropic's Fable model for K3's development.
The current administration (at the time of recording) weighing in on the debate about banning Chinese open-source models.
Mentioned as a friend of the show who does not support banning Chinese AI models.
A political scientist whose work is referenced to elevate the discussion on the geopolitical implications of banning Chinese AI models.
A U.S. senator whose concerns about wealth concentration are contrasted with the democratizing effect of open-source AI.
The leader of China, whose potential response to a U.S. ban on Chinese AI models is considered for its escalatory implications.
A technology analyst and writer whose blog discussed the cost numbers of Kimmy K3, finding it not significantly cheaper.
A basketball player used as an analogy for 'flopping' or exaggerating to draw a foul, comparing it to AI companies seeking government intervention.
CEO of OpenAI, mentioned as going to Washington to discuss GPT-6.0 and publicly stating OpenAI has its 'mojo back'.
Chair of the Federal Trade Commission, suggested sarcastically that Anthropic should hire her, implying a focus on antitrust or regulatory issues.
A U.S. senator whose concerns about wealth concentration are contrasted with the democratizing effect of open-source AI.
An investor who appeared on the podcast previously, mentioned for his stance on AI token growth, despite the commoditization trend.
Mentioned in the context of 'Elon's Web Service' and Grok, hinting at his involvement in AI and infrastructure.
A state cited as an example (along with Texas and Florida) that has successfully increased housing supply by being less restrictive with building regulations.
A state cited as an example (along with Florida and Nevada) that has successfully increased housing supply by being less restrictive with building regulations.
A country mentioned where the removal of rent control led to a decrease in rents.
A city mentioned as an example of failed socialist experiments, implying that its issues could be extrapolated to New York City.
The site of a rental 'ripoff' hearing led by Mayor Zohan Mani.
A city mentioned as demonstrating that relaxing permitting constraints leads to more housing units and lower rents.
A state cited as an example (along with Texas and Nevada) that has successfully increased housing supply by being less restrictive with building regulations.
More from All-In Podcast
View all 434 summaries
90 minCan the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters
50 minFormer Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe Coding
52 minThe Trillion-Dollar Industries AI Is Disrupting: Voice, Law & the End of the Billable Hour
103 minMore Trillion Dollar IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free