Summary
No publicly traded tickers discussed. All major companies referenced (Anthropic, OpenAI, xAI) are private. Episode focuses on AI policy/regulation, inflation data, and California election integrity.
Theses (episode spine)
- Anthropic’s Fable 5 model tops benchmarks but sparked major developer backlash by silently downgrading users it profiles as AI researchers, retaining all prompts for 30 days with no opt-out even for enterprise zero-data-retention customers, and hiding the degraded capability while still charging full price — the narrow walkback was only to disclose downgrades, not to stop them.
- Sacks argues Anthropic is engaged in a sophisticated regulatory capture campaign: simultaneously surveilling users and nerfing models for AI researchers while CEO Dario Amodei calls for a new FAA/FDA-style regulatory agency — the real target is open-source models, which cannot be regulated, giving Anthropic a duopoly.
- Freeberg warns that Anthropic’s biosafety restrictions on legitimate genomics and gene-editing research are forcing companies to adopt Chinese open-source models (which are superior to American open-source alternatives), handing China an unfair competitive advantage in biotech, materials science, and AI.
- Chamath argues that if you want to support open-source AI at frontier scale, the capital barrier is now roughly $100B per gigawatt of compute — he has 2 GW zoned in Arizona and is bidding on another GW, believing the community may be dragged into building 3 GW of open-source-accessible compute.
- On Bernie Sanders’ proposal to impose a one-time 50% tax on AI company equity into a sovereign wealth fund: Sacks has sympathy for the politics given AI CEOs predicted 50% job loss, but opposes confiscation; Freeberg proposes restructuring Social Security into an equity-investing sovereign wealth fund; Chamath says if negotiating he would end up owning 75% given AI’s high marginal compute cost.
- Sacks and Freeberg both reject the AI job-loss narrative empirically: May jobs report showed 172,000 new jobs (double expectations), software developer jobs at a three-year high, 4.3% unemployment; Freeberg says AI is being used overwhelmingly on the revenue side (expanding products) not the cost-cutting side.
- May CPI printed 4.2% year-over-year (highest since April 2023) and PPI 6.5% year-over-year (highest since end-2022), driven largely by the Iran war energy shock; Polymarket pricing Fed hike at 49% this year; ECB raised rates a quarter point Thursday for the first time since September 2023.
- On the LA mayoral primary: Freeberg and Sacks argue California’s combination of universal mail-out ballots, no voter ID, unlimited ballot harvesting, no chain of custody, and near-impossible audits has converted elections into appointments; the statistical shift (Pratt winning in-person 35% but losing post-election mail-ins) is cited as evidence, though the hosts disagree on whether acts are illegal fraud or legal-but-corrupt exploitation of loopholes.
- Thomas Laffont data presented at All-In Liquidity showed odds of further valuation multiples increase with scale: 8% of unicorns reach decacorn, 13% of decacorns reach centicorn, 31% of centicorns reach trillion-dollar market cap — hosts interpreted this as validating concentrated positions in the largest companies.
Topics discussed
Anthropic Fable 5 Backlash — Surveillance, Silent Downgrading, and Regulatory Capture
Summary: Anthropic released Fable 5, a top-benchmarking model that retains all user prompts for 30 days with no exceptions, even for enterprise zero-data-retention customers. It silently downgraded users it profiled as doing AI or chip-design research, charged them full price, and in some cases rewrote their prompts in the background without disclosure. After developer backlash, Anthropic walked back only the disclosure requirement, not the downgrading itself. Sacks framed this as regulatory capture: Anthropic surveils users while pushing for a government agency to regulate AI, with open-source models as the real target.
Speaker views:
- Chamath: Anthropic has shown their hand — they will increasingly evaluate prompts before generating output, creating censorship risk for individuals and a governance/single-point-of-failure risk for enterprises. The lack of KYC implementation despite the fear-mongering is the tell of an ulterior regulatory agenda.
- Freeberg: Anthropic’s biosafety restrictions already block legitimate genomics work such as RNA guide design and genetic construct design, forcing his company and others to use Chinese open-source models which are currently superior to US open-source alternatives, handing China a structural AI advantage.
- Sacks: The key violation is that Anthropic retained data even for enterprise zero-data-retention customers, nerfed models without disclosure, and is simultaneously calling for a new regulatory agency — the goal is to sandbag open-source competitors who cannot be regulated.
- Jason: There is a steelman for Dario — if you genuinely believe you built a dangerous model, cautious rollout with monitoring is defensible — but the lack of KYC, the hypocrisy of blogging about recursive self-improvement risks while hiring to run it, and downgrading everyday questions undermine the case.
Potential impact: Hosts argue model restrictions will accelerate adoption of Chinese open-source models across US startups and enterprises, eroding US competitive advantage in biotech and AI; a regulatory regime shaped by Anthropic’s preferences could entrench a closed-model monopoly or duopoly.
Nationalizing AI — Bernie Sanders’ Sovereign Wealth Fund Proposal
Summary: Senator Bernie Sanders proposed a one-time 50% tax on the stock of the largest AI companies (OpenAI, Anthropic, xAI) to fund an American AI Sovereign Wealth Fund with public voting rights and board seats. The hosts debated whether confiscation is justified given AI CEOs have repeatedly predicted mass job loss and trained on publicly available human knowledge. Freeberg proposed an alternative: restructure Social Security into an equity-investing sovereign wealth fund.
Speaker views:
- Sacks: Opposes the Sanders proposal as straightforward confiscation and a terrible precedent, but has sympathy for the political logic because AI CEOs have themselves told the public that AI will destroy 50% of entry-level knowledge-worker jobs, making a public ownership demand a natural political response.
- Freeberg: Opposes wealth taxes as asset seizures; proposes restructuring the Social Security trust fund — currently a single $4 trillion Treasury certificate — to allow equity investments in American companies including AI, turning it into an account-based sovereign wealth fund where every citizen holds shares.
- Chamath: Calls Freeberg’s idea genius but politically impossible; argues that AI’s high marginal compute cost (unlike zero-marginal-cost internet platforms) gives government genuine leverage, and that if he were negotiating he would end up owning 75% of these companies given the infrastructure dependency.
- Jason: Says the AI companies asked for equity seizure by publicly describing themselves as an existential risk; notes Sam Altman appears open to some form of public participation.
Potential impact: A 50% equity tax or mandatory sovereign fund stake would fundamentally alter the capital structure of the largest AI companies and could deter future private AI investment. Polymarket gives Anthropic IPO odds at 83% before end of 2027, OpenAI at 48%.
Open-Source AI and Compute Infrastructure
Summary: The hosts discussed the economics of building open-source frontier AI, with Chamath disclosing he owns 2 GW of zoned and approved data-center land in Arizona and has put in an offer for another GW elsewhere. He argued the community may have to build 3 GW of compute specifically to sustain open-source model access because current megawatts are overwhelmingly directed to closed frontier labs. Freeberg noted that open-source genome language models (e.g., the Ark Institute model backed by the Collisons) are already in active use and cannot be un-published.
Speaker views:
- Chamath: A gigawatt of data center capacity now costs roughly $100 billion — a 20x increase since he started the Arizona project — making 3 GW of open-source compute a $300 billion undertaking that no single private investor can fund; he is actively pursuing additional capacity because he has come to the conclusion the community has to do it.
- Freeberg: Open-source models, once published, cannot be regulated away — they are like printed books; the Ark Institute genome language model is a working example of community-funded open-source AI already delivering competitive advantage for specialized scientific work over closed models.
- Jason: Offered to seed-invest in any team wanting to start an open-source frontier model company.
Potential impact: If regulatory capture restricts closed frontier models without viable open-source alternatives, the hosts argue the US risks ceding AI leadership to China, which already offers superior open-source models that US companies are actively adopting.
May Inflation — CPI 4.2%, PPI 6.5%, ECB Rate Hike
Summary: May CPI came in at 4.2% year-over-year, the highest since April 2023, and PPI at 6.5% year-over-year, the highest since end-2022. The ECB raised rates by a quarter point Thursday, its first hike since September 2023. Polymarket put the probability of a Fed rate hike in 2026 at 49%, up from under 10% before the Iran war. The NASDAQ was up 2.5% on the print day, which the hosts interpreted as the market pricing in a geopolitical resolution to the Iran conflict.
Speaker views:
- Freeberg: The print is an energy blip from the Iran war layered on top of structurally out-of-control government spending; with a Kevin Warsh-led Fed, overnight rates going north of 5.5% to 6% is not unforeseeable.
- Chamath: China’s strategic energy reserves have kept oil sub-$100 and dampened the inflationary spike, but if China needs to return to spot markets for an extra 3 million barrels per day, oil could reach $150-$200 a barrel with severe downstream CPI impact; the PPI number should be an alarm to seek an off-ramp from the Iran situation sooner rather than later.
- Sacks: The print was largely in line with market expectations, which is why equities are up; a surprise to the upside would have hit markets harder.
Potential impact: A sustained energy shock from the Iran war could push oil to $150-$200 per barrel and CPI toward or past 5%, prompting Fed rate hikes that would disproportionately pressure tech stocks.
LA Mayoral Primary — Election Integrity and California Ballot Laws
Summary: Statistical anomalies in the Los Angeles mayoral primary: Spencer Pratt led in-person voting 35% to Raman’s 26%, but post-election-day mail-in ballots flipped dramatically with Raman rising to 37% and Pratt dropping to 19%. The discussion covered California laws enabling unlimited ballot harvesting (AB 1921), universal mail-out ballots, no voter ID requirement, and recent laws making audits harder.
Speaker views:
- Freeberg: There is no outright illegal fraud — the system is operating as designed by a series of California laws that in aggregate convert elections into appointments; Congress must pass federal voter ID and chain-of-custody laws.
- Sacks: Whether or not individual acts are technically legal, the statistical swing in post-election mail-in ballots is impossible to explain legitimately; one person (Brenda Lee Brown Armstrong) has already pled guilty to paying people on Skid Row to vote.
- Chamath: California is a one-party state where the political machine has shaped laws to enforce a monopoly; the laws have made legal what would otherwise be an illegality.
- Jason: The statistics are alarming but Trump’s DOJ has the ability to investigate; federal voter ID legislation is the clean fix, and both parties bend election rules.
Venture Capital Return Distribution — Thomas Laffont Data from All-In Liquidity
Summary: Thomas Laffont presented data at the All-In Liquidity event showing that odds of achieving further large valuation multiples increase with scale: 8% of unicorns reach decacorn, 13% of decacorns reach centicorn ($100B+), and 31% of centicorns reach trillion-dollar market cap. Jason extrapolated this pattern and suggested roughly 60% of trillion-dollar companies might reach $10 trillion in the coming years.
Speaker views:
- Sacks: The data counterintuitively shows that getting from $100B to $1T is statistically more likely than getting from $10B to $100B — a finding that should fundamentally change how capital allocators think about concentration at the top of the venture distribution.
- Jason: If the pattern continues to hold, seizing 10% of 30 trillion-dollar companies would pay off approximately 2% of the US national debt — making the sovereign wealth fund math more credible than it sounds.