Podcast
Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California's Broken Elections
All-In with Chamath, Jason, Sacks & Friedberg
- Fable 5’s Hidden Surveillance Creates Censorship Risk
- Anthropic’s Fable 5 retains all prompts/outputs for 30 days and profiles users to decide capability access.
- Chamath warns this creates censorship risk and enterprise governance danger because companies can be downgraded without clear notice. (Time 0:01:34)
- Underwrite AI Control With Diverse Models
- Companies should underwrite control and diversify AI suppliers to avoid single-point-of-failure risk.
- Chamath recommends governance approaches and multiple models so enterprises keep control over who learns from their data. (Time 0:03:55)
- Genomics Team Lost Capabilities After Model Nerfs
- David Friedberg describes using LLMs for genomic work and being blocked from tasks like RNA guide design.
- He says restrictions forced his company to pivot to open source Chinese models and build internal genome models instead. (Time 0:05:15)
- Nerfs Could Accelerate Chinese Open Source Adoption
- Friedberg warns heavy guardrails on proprietary models push companies toward open source Chinese models.
- He notes US open source alternatives lag Chinese models, creating an economic and national-security disadvantage. (Time 0:07:21)
- Hidden Downgrades Create AI Haves And Have Nots
- Anthropic downgraded users doing frontier research without disclosure and even rewrote prompts, creating an AI haves/have-nots dynamic.
- David Sacks highlights outrage that downgrades were hidden in a long policy document. (Time 0:10:12)
- Regulate Outputs Not Model Access
- Focus regulation on dangerous outputs instead of preemptively banning models; codify existing synth DNA screening and record keeping.
- Sacks and Friedberg support mandatory nucleic acid screening as a targeted downstream guardrail. (Time 0:33:31)
- Political Backlash Makes AI A Public Resource Debate
- Bernie Sanders proposed a one-time 50% tax to form an AI Sovereign Wealth Fund, arguing AI companies trained on public knowledge.
- David Sacks and others acknowledge political pressure arises because companies told the public AI threatens jobs. (Time 0:38:26)
- Convert Social Security Into A Sovereign Equity Fund
- Reform Social Security into an equity-owning sovereign fund to give citizens ownership in AI gains.
- David Friedberg proposes moving the trust to an account-based system that can buy equities in transformative companies. (Time 0:45:03)
- AI Has Real Marginal Costs Unlike Social Platforms
- Chamath notes AI’s economics differ from internet platforms because every marginal AI user consumes real compute and energy.
- That high marginal cost is an argument for government stake in critical AI infrastructure. (Time 0:53:30)
- Inflation Spike Tied To Energy Shock And Spending
- May CPI and PPI prints rose to multi-year highs, partly due to energy effects from the Iran war and fiscal overspending.
- David Friedberg warns higher rates could return, predicting possible 5.5–6% overnight rates under current policy. (Time 1:05:43)
- California Laws Create Opportunities For Post Election Swings
- California’s election rules (ballots to all, unlimited ballot harvesting, relaxed registration) create incentives and pathways for large post‑election vote swings.
- David Friedberg calls this legal design that can make elections functionally appointment-based. (Time 1:13:44)
- Operational Flaws Make Mail Ballots Vulnerable
- David Sacks outlines operational vulnerabilities: dirty voter rolls, weak signature checks, no voter ID, and lax chain of custody for mailed ballots.
- He argues those weaknesses plus recent laws reduce auditability and raise fraud incentives. (Time 1:19:54)
- Fix Elections With Voter ID And Clean Rolls
- Congress should require voter ID and tighten registration rules to restore one-person-one-vote integrity.
- David Friedberg urges federal law to prevent gym-card registrations and allow meaningful audits and chain-of-custody checks. (Time 1:29:50)