All-In PodcastE129: Sam Altman plays chess with regulators, AI's "nuclear" potential, big pharma bundling & more
CHAPTERS
- 0:00 – 5:14
Reddit performance reviews: roasting the hosts and audience feedback as a bit
Jason introduces a new segment where each host reads blunt Reddit critiques aloud. The group riffs on “360 reviews,” ego, and the show’s dynamic while turning the insults into comedy.
- •Jason debuts “Reddit Performance Reviews” as a new show feature
- •Friedberg, Sacks, Chamath, and Jason each read harsh audience comments
- •Banters about 360 reviews, elite performers, and outsourcing judgment to Reddit
- •Jokes about renaming the podcast based on a “binding” Reddit poll
- 5:14 – 6:25
All-In Summit 2023 logistics: tickets, venue capacity, and Chamath’s wine budget
They briefly promote the All-In Summit, explaining sold-out general admission and limited VIP availability. Chamath jokes about the wine budget and appoints “Sommelier Josh,” keeping the tone light before moving into news.
- •General admission sold out; scholarships paused due to demand
- •Royce Hall venue details and maintaining event quality
- •VIP tickets still available; quick promotional update
- •Chamath riffs on the wine budget and “curating” wine
- 6:25 – 13:33
AI Senate hearing kickoff: Altman’s proposed AI agency, licensing, and regulatory capture concerns
The besties react to Sam Altman’s Senate testimony and his call for a new AI regulator with licensing authority. They frame it as either prudent risk management or strategic “moat building” by incumbents.
- •Altman proposes a new AI oversight agency with licensing and audits
- •Debate over whether this is sensible safety policy or regulatory capture
- •Sacks argues Altman “played ball” to shape rules and protect incumbency
- •Senators’ stated harms (misinformation, elections, copyright) vs existing laws
- 13:33 – 24:27
Open-source reality check: edge models, Hugging Face leaderboards, and why auditing AI is impractical
Friedberg argues regulation is structurally difficult because models are shrinking, commoditizing, and moving to the edge. Jason and Sacks expand with examples from Hugging Face and open-source leaderboards to show enforcement challenges.
- •Trend: smaller models running on phones/edge devices and proliferating rapidly
- •Open-source ecosystems (Hugging Face leaderboards) accelerate performance and diffusion
- •Regulating ‘algorithms inside apps’ becomes a massive audit problem (US and EU AI Act)
- •Sacks: existing laws already cover many harms; licensing would kill permissionless innovation
- 24:27 – 36:15
‘Nuclear’ potential debate: AGI risk, cyber/biochem fears, KYC for GPU clusters, and slowing the ship
Chamath pushes a high-stakes analogy—AI as nuclear weapons—arguing rare but catastrophic misuse warrants controls. The group debates what can be regulated (training compute via KYC) versus what’s already out of the bag (weights, open models).
- •Chamath: 99.9% beneficial, 0.1% existential/catastrophic—needs guardrails
- •Explainer: model weights vs model structure; LLaMA weights leak increases risk
- •Proposal: KYC/identity requirements for large GPU cluster training runs (cloud gatekeeping)
- •Sacks: species-level risk is mainly AGI, but licensing regimes are political control mechanisms
- •Examples of malicious use: cyber exploits, automated hacking/‘auto-GPT’ style agents
- 36:15 – 43:29
Insider Washington vibes: White House AI summit skepticism and fear of tech-driven job displacement
Sacks shares secondhand impressions from Capitol Hill and a White House AI meeting described as “rabidly negative.” The group argues policymakers may be motivated by fear—especially of job disruption—more than opportunity.
- •Reported mood at White House AI summit: pessimism, desire to ‘stop the hot new thing’
- •Critique of policymakers’ technical understanding and governance competence
- •Discussion: job displacement vs productivity gains and new-company creation
- •Concern about repeating social media-era playbooks (disinformation narratives → censorship machinery)
- 43:29 – 48:20
Twitter’s new CEO: Linda Yaccarino’s ad-sales focus and how she can coexist with Elon
They evaluate Elon Musk hiring Linda Yaccarino as Twitter CEO, emphasizing her advertising expertise. The consensus: success depends on complementary roles and letting Elon lead product and speech policy while she repairs advertiser relationships.
- •Strategic rationale: Twitter depends on ads; Elon dislikes ad sales; Linda excels at it
- •Predicting a 6–9 month evaluation window for performance
- •Culture-war scrutiny: attacks from both left and right based on affiliations/follows
- •Sacks: model could mirror Musk/Shotwell at SpaceX—business vs product split
- 48:20 – 57:20
Amgen–Horizon blocked by FTC: bundling as the real antitrust issue, not deal size
The group debates Lina Khan’s FTC lawsuit to stop Amgen’s $27.8B Horizon acquisition. Friedberg finds the FTC’s theory compelling—Amgen’s portfolio enables bundling tactics that raise prices and block competition—while Chamath worries about chilling biotech funding incentives.
- •Chamath: pharma R&D economics depend on M&A to fund risky clinical development pathways
- •Friedberg: FTC case focuses on negotiated multi-drug bundling with payers/insurers as anti-competitive
- •Sacks: regulate tactics (bundling) rather than blanket-blocking mergers; preserve IPO/M&A outcomes
- •PBMs highlighted as major drivers of drug pricing dynamics; FTC also probing PBMs
- 57:20 – 1:02:30
ChatGPT product momentum: link summarization, browsing-like behavior, and mobile app launch
In a quick tangent, Jason describes ChatGPT summarizing a legal PDF via a link without manual upload, surprising the group. They note product velocity, compare to Bard, and react to the official ChatGPT app launch as a distribution accelerant.
- •ChatGPT appears to fetch and summarize content from a shared URL directly
- •Speculation that browsing is being integrated beyond plugins
- •ChatGPT iOS app launch expected to expand user base and usage dramatically
- •Call for comparisons between Bard and ChatGPT summaries
- 1:02:30 – 1:07:30
Apple’s AR headset rumors: early prototype strategy, platform-shift bets, and AI+AR ‘Terminator mode’
They discuss reports Apple will unveil a $3,000 AR/VR headset and why it departs from Apple’s usual polished mass-market launches. The conversation shifts to the long-term payoff—lightweight glasses paired with AI that augment reality with contextual information.
- •Apple rumored to ship a developer-first, expensive, imperfect first version (external battery, ‘ski goggles’)
- •Sacks: iterating early can work (Apple Watch improved over generations) and pressures Meta
- •Debate over real use cases vs novelty; whether consumers will pay for eye-based notifications
- •AI+AR concept: real-time identity/context recall, navigation, personalized overlays
- 1:07:30 – 1:10:56
Housing affordability collapse: Gallup sentiment lows, rate shock, and frozen mobility
They interpret a Gallup survey showing record-low belief that it’s a good time to buy a home. Sacks ties it to higher mortgage rates and weakened confidence, arguing it freezes home turnover and reduces geographic mobility and opportunity.
- •Only 21% say it’s a good time to buy—historic low sentiment
- •Higher mortgage rates (near ~7%) crush affordability compared to recent 3% era
- •Lock-in effect: owners hesitate to sell and give up low-rate mortgages
- •Reduced mobility can increase dissatisfaction and distort price discovery
- 1:10:56 – 1:15:29
Commercial real estate alarm: San Francisco office vacancy, ‘pretend and extend,’ and banking risk
Sacks shares broker/Blackstone anecdotes and bleak SF office stats, arguing the fallout will hit banks as collateral values reset. Jason contrasts NYC’s street-level vibrancy with SF’s emptiness, then they discuss how demand concentrates in newer, amenity-rich buildings.
- •SF office market cited at ~90M sq ft with ~35% vacancy; sublease supply rising
- •AI startups can’t absorb the surplus; landlords lack capital for tenant improvements
- •‘Pretend and extend’ loan restructures delay recognition of losses but don’t solve insolvency
- •Flight to quality: tenants want best locations/buildings; older towers risk becoming ‘zombies’
- 1:15:29 – 1:24:09
Self-defense tragedies in SF and NYC: policing, media framing, and mental health as root cause
They compare two high-profile incidents—an SF shoplifter shot by security and the NYC subway restraint death—arguing public disorder creates flashpoint confrontations. The discussion broadens to prosecutorial discretion, media narratives, and systemic failures in mental health services.
- •Contrasting prosecutorial responses in SF vs NYC and the politics around them
- •Debate over media portrayal and omission of background details in the NYC case
- •Historical parallels: Bernie Goetz era and Guardian Angels as responses to lawlessness
- •Policy turn: expand mental health facilities/services to prevent repeat tragedies
- 1:24:09 – 1:28:04
Soros and progressive DAs: prosecutorial discretion as de facto lawmaking and broader political influence
Jason asks what motivates Soros-backed DA campaigns; Sacks argues the mechanism is changing enforcement priorities without legislation. He expands to Soros’s Open Societies work internationally, framing it as destabilizing interference viewed as regime-change operations.
- •Claim: Soros influences cities via DA elections, shifting policy through prosecution choices
- •Argument that non-prosecution patterns create perceived lawlessness
- •Discussion of Open Societies Foundation and ‘color revolution’ accusations abroad
- •Invitation idea: bring Soros or his son on for an interview; episode wrap-up
