Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California's Broken Elections

Summary

The hosts open with a detailed critique of Anthropic’s Fable 5 release, praising its benchmark performance while accusing the company of broad surveillance, hidden downgrades, and overly aggressive safety filters that can flag benign work. They argue that these guardrails push serious users toward open-source and local models, while the real policy question should be how to regulate harmful outputs like weapons and bioweapons rather than restricting access to model capability itself. The conversation then expands into nationalization and wealth-sharing proposals for AI, including Bernie Sanders-style taxation, Social Security as an equity-owning sovereign wealth fund, and the claim that AI’s real infrastructure bottleneck is power and capital. In the macro segment, they interpret hot CPI/PPI prints as evidence that inflation is reaccelerating and that rates could stay higher for longer. The episode closes with a combative discussion of California election mechanics, where the hosts argue that ballot harvesting, weak ID rules, and audit limitations make the system vulnerable to “legal fraud” and legitimacy collapse.

Notes

• Fable 5 is presented as a top-tier benchmark model, but at a much higher token cost than Opus 4.8.
• Anthropic is criticized for storing prompts for at least 30 days and for downgrading users suspected of frontier AI research without telling them.
• Chamath says this creates censorship risk for individuals and a business “non starter” for companies.
• Friedberg says Anthropic’s restrictions are already pushing genomics work toward open-source models run locally.
• He adds that the strongest open-source models today are Chinese, which he flags as a strategic concern.
• Sacks says Anthropic keeps every prompt and output for 30 days, including enterprise users with zero-retention expectations.
• He says the backlash is also about silent downgrades that create AI “haves and have nots.”
• Anthropic later appears to walk back disclosure of downgrades, but not the downgrades themselves.
• Chamath worries Anthropic could favor partners by shaping information flow for competing corporations.
• He also says the audit trail may be missing, making manipulated model runs hard to reconstruct.
• Chamath says frontier AI infrastructure has become a huge capital moat, with a gigawatt now costing about $100 billion.
• Friedberg points to Evo 2 as an open-source genome language model trained on broad genomic data.
• He says the model is useful in plant breeding by scoring DNA letter sequences for likely good or bad variants.
• Jason says Dario’s safety worldview may be extreme, but many AI researchers likely share it.
• Friedberg analogizes AI risk to nuclear technology: the same breakthrough can enable both bombs and abundant energy.
• He argues regulation should target weaponization outputs, not access to the tools themselves.
• Sacks says major AI labs support mandatory nucleic acid screening and record-keeping for DNA synthesis.
• Friedberg says DNA/RNA vendors already do this screening and can codify it without slowing research.
• Friedberg says open-source models cannot be turned off once released, likening them to books and Xerox machines.
• Jason says Fable 5 blocks cybersecurity and biology prompts, including a nuclear bomb follow-up that switched the user to Opus 4.8.
• Friedberg says these blocks will push users to competing models.
• The hosts then discuss Bernie Sanders’ proposed 50% AI tax and sovereign wealth fund framing.
• Sacks rejects confiscation but says he understands the political impulse behind a public stake.
• Sacks argues public anger comes from AI leaders warning of 50% job losses while using humanity’s knowledge as training data.
• Sacks says the May jobs report does not support an AI-apocalypse labor story.
• Friedberg argues AI’s main effect is productivity expansion on the revenue side, not broad labor destruction.
• He says his own team wants 15 more engineers because AI has expanded the amount of work worth doing.
• Sacks distinguishes Elon’s abundance thesis from Dario’s near-term job-loss forecast.
• He says Dario specifically predicted 50% job loss for entry-level knowledge workers within one to five years.
• Sacks says the public now associates AI with job loss, making wealth-seizure rhetoric politically understandable.
• He says Anthropic’s anti-RSI blog post looks hypocritical next to its Andrej Karpathy hire.
• Sacks suggests the blog also helped justify depriving users of AI, ML, and chip-design research capabilities.
• Chamath says AI inference has non-zero marginal cost because every user consumes GPU, power, and memory.
• He argues that difference supports a case for government ownership stakes in AI firms.
• Chamath says the real constraint is power delivery, not the labs themselves.
• The discussion also includes venture scaling odds from Thomas Lafont’s presentation.
• Sacks gives the progression rates as 8% from unicorn to decacorn, 13% from decacorn to centicorn, and 31% from centicorn to trillion-dollar company.
• Jason frames the May CPI/PPI print as hot and links it to higher Fed-hike odds.
• Friedberg says the Iran war created an energy blip and that rates could move above 5.5% to 6%.
• Chamath says China has been smoothing energy demand and that oil could spike sharply if China returns to the spot market.
• The California segment begins with skepticism about the LA mayoral race and whether the election is legitimate at all.
• Friedberg answers that Californians are effectively assigned overseers rather than choosing representatives.
• He then cites large turnout differences by ballot type as suspicious.
• Friedberg says California normalized unlimited ballot harvesting, universal ballot mailing, and weak identity checks.
• Sacks says California sends ballots to every registered voter, voter rolls are dirty, and signature verification is weak.
• He argues California’s audit regime and fraud environment make abuse plausible, not far-fetched.
• Chamath says California is effectively a one-party autocracy with rules designed to preserve monopoly control.
• Chamath proposes electing Steve Hilton and using a day-one state of emergency as a “break the glass” strategy.
• Friedberg says Congress should require ID, citizenship, and proper registration for voting.
• Sacks says the result is either illegal fraud or legal fraud created by loopholes.
• Near the end, Jason says there is no evidence of fraud yet, but asks Trump and DOJ to investigate ballot harvesting.