All-In PodcastGPT-4o launches, Glue demo, Ohalo breakthrough, Druck's Argentina bet, did Google kill Perplexity?
CHAPTERS
- 0:00 – 15:00
Cold Open, Vegas Degeneracy, and Baccarat Philosophy
The episode opens with birthday jokes, Vegas trip stories, and a long riff on baccarat as the purest form of degeneracy. The besties contrast it with craps and mock how even highly intelligent people invent rituals to feel in control of pure coin‑flip odds.
- •Lighthearted banter about Chamath’s child’s birthday and Phil Hellmuth’s oversharing in Vegas.
- •Descriptions of high‑stakes baccarat and how players convince themselves technique matters when it’s effectively high‑card.
- •Observation that smart people can irrationally obsess over meaningless patterns (baccarat sweat, card bending) to feel agency in random outcomes.
- 15:00 – 28:00
Programming Notes: Sam Altman Interview and ‘We’re Not Journalists’
They debrief why their Sam Altman interview aired just before the GPT‑4o launch and why he couldn’t discuss it. This leads to a clear statement that All‑In is not a journalistic outlet and won’t do gotcha interviews, prioritizing curiosity-driven conversations over adversarial grilling.
- •Friedberg explains the scheduling: Altman had planned a big announcement, delayed it, and couldn’t talk about GPT‑4o when he came on.
- •The hosts acknowledge it was a ‘lesson learned’ about timing big product launches and guest slots.
- •Chamath and Sacks stress they have day jobs and aren’t journalists; guests come for candid conversations, not adversarial press conferences.
- 28:00 – 51:00
GPT‑4o (Omni): Architecture Shift and New Use Cases
The besties analyze OpenAI’s GPT‑4o launch, focusing on its multimodal, real‑time capabilities and what it signals about the future of model architecture. They highlight demos like real‑time translation, math tutoring via desktop capture, and how faster, cheaper models change product and pricing strategy.
- •GPT‑4o is ‘Omni’: it processes audio, text, images, video, and desktop context simultaneously.
- •Friedberg argues we’re moving from monolithic models to interconnected, continuously updated, specialized models.
- •Stanford MMLU data shows GPT‑4o edging GPT‑4, contradicting early ‘degraded model’ narratives, though Claude 3 Opus still tops that particular benchmark.
- •Chamath suggests consumer growth in ChatGPT has plateaued; looky‑loos dropped off, and real use cases must drive the next wave.
- •Sacks predicts B2C subscriptions aren’t the best business; OpenAI’s bigger opportunity is B2B and platform economics via apps built on top.
- 51:00 – 1:10:00
Glue: An AI‑Native Slack Killer Without Channels
David Sacks demos Glue, his AI-native enterprise chat product that eliminates channels in favor of topic threads and deeply embeds GPT‑4o. They show Glue AI answering questions about All‑In episodes, profiling each host’s personality, and drafting a sourced invitation letter based on podcast transcripts.
- •Glue replaces Slack’s channels with topic-based threads, reflecting how ChatGPT organizes conversations.
- •Glue AI uses GPT‑4o plus a RAG layer (via Raggy) to query transcripts and enterprise data repositories (e.g., ‘Episodes’ group, dealflow).
- •Examples: identifying Sacks’ most-discussed countries, summarizing his Ukraine stance with citations, and drafting a Lina Khan invitation letter referencing past praise.
- •Future vision is ‘promptless’ AI: bots that proactively chime in (e.g., sales reminders) when they detect relevance, requiring better intent modeling.
- •Chamath and JCal praise Glue’s minimalism and lack of ‘random’ channels that waste time and make Slack feel like the job instead of a tool.
- 1:10:00 – 1:25:00
Ohalo’s ‘Boosted Breeding’: Doubling Plant DNA to Supercharge Yields
Friedberg reveals Ohalo’s breakthrough ‘boosted breeding’ technology that alters plant reproduction so both parents pass 100% of their genes to offspring, creating controlled polyploids. The data shows dramatic yield increases in model plants and especially potatoes, with huge implications for global food security, new seed industries, and crop adaptation to harsh environments.
- •Normal plant reproduction halves DNA contributions from each parent, slowing breeding and making it hard to stack traits like drought and disease resistance.
- •Ohalo uses proteins to switch off meiotic halving, so each parent passes its full genome; offspring have double the chromosome sets (polyploidy).
- •In Arabidopsis, boosted plants are visibly larger with ~40%+ seed yield increases; in potatoes, two low‑yield parents (33g, 9g) produced a 682g offspring.
- •Many crops (wheat, potatoes, strawberries) already exhibit natural polyploidy; Ohalo industrializes and directs this phenomenon.
- •Boosted breeding also enables true, stable seed in crops like potatoes that currently must be propagated via cut tubers, reducing disease, logistics cost, and giving farmers ~20% revenue savings.
- •Friedberg envisions adapting staple crops to new climates (hot, dry, sandy regions), addressing malnutrition in regions like parts of Africa and South Asia.
- •Ohalo has spent >$50M over five years, holds patents but sees the main moat as compounding genetic advantage and trade secrets, not lawsuits.
- 1:25:00 – 1:42:00
Global Food, Geopolitics, and the Business of Seeds
The conversation zooms out to food security and geopolitics: how boosted breeding could change where crops grow and undercut reliance on traditional breadbaskets. They also touch on business models (selling seed vs. farming), IP moats, and consumer-facing traits like fruit flavor and size.
- •Boosted crops could help grow staples in currently marginal regions (Somalia, Sudan, Yemen, Afghanistan), shifting geopolitical leverage away from a few breadbaskets.
- •Friedberg emphasizes partnering with existing seed companies in some crops and building proprietary seed businesses in others (e.g., potato).
- •Moat strategy is continuous genetic improvement and farmer loyalty, not primarily IP litigation.
- •Side discussion: Driscoll’s ‘Sweetest Batch’ berries and how breeding targets traits like sweetness or size; Friedberg’s team includes ex‑Driscoll’s breeders.
- •Chamath and JCal immediately angle to invest, underscoring perceived scale and defensibility of the technology.
- 1:42:00 – 1:54:00
AI, Offshore Talent, and the Deflation of Startup and Operating Costs
The besties reflect on how AI and global labor arbitrage are making companies cheaper to start and run. JCal describes his biggest-ever seed bet into Athena, an offshore EA/operations platform, and introduces his ‘Automate, Deprecate, Delegate’ framework as a way to systematically reconfigure work around AI and lower-cost talent.
- •Athena pairs executives with highly trained offshore operators (e.g., in the Philippines), often at a quarter to a third of U.S. cost, then layers in AI tools.
- •Calacanis uses Athena to ‘80/90’ his investment firm—offloading repetitive screening and diligence so high‑paid staff move up the value chain.
- •His ADD framework: Automate (use AI), Deprecate (stop low‑value work), Delegate (offload to cheaper human labor).
- •Startups now routinely request $500K–$1M instead of $3M+ for products that once required far more capital, making many niche ideas economically viable.
- •Chamath notes that this mirrors past deflation in compute; now it’s happening in headcount and OPEX—shrinking orgs while raising per‑person leverage.
- 1:54:00 – 2:04:00
Druckenmiller’s Argentina Bet, Milei’s Austerity, and Bidenomics Critique
They analyze Stanley Druckenmiller’s big bet on Argentina after Javier Milei’s Davos speech, highlighting Milei’s radical spending cuts and rapid return to surplus. This segues into a critique of U.S. fiscal policy under Biden, arguing that massive deficit spending into a hot economy is fueling inflation and crowding out innovation.
- •Druckenmiller used Perplexity to surface the five most liquid Argentinian ADRs, bought them all immediately, then sized up after research—‘invest, then investigate’ à la Soros.
- •Milei turned a 4–5% primary deficit into ~3% surplus by cutting entitlements (e.g., social security) and slashing government spending, even as GDP dipped sharply.
- •Sacks stresses the basic macro equation: you can’t run deficits forever without eventual inflation or default, reserve currency status only buys time.
- •He criticizes ‘Bidenomics’ for deploying depression-style stimulus into a strong economy, forcing high interest rates that hurt mortgages, car loans, and investment.
- •They note the conflicting posture: the Fed is braking with high rates while Congress and the White House are flooring the gas with deficit spending.
- 2:04:00 – 2:14:00
Precision vs. Accuracy in Investing and Network-Effect Bets
Friedberg contrasts ‘precision’ investing (deep micro-diligence that can miss macro trends) with ‘accuracy’ investing (getting the big trend right and being patient). The besties share war stories about missing obvious network effects or over-analyzing incumbents, reinforcing the value of betting on momentum and founder quality over perfect spreadsheets.
- •Precision: detailed margin and P&L analysis can still miss existential threats (e.g., buying Macy’s while ignoring Amazon’s e‑commerce trend).
- •Accuracy: betting on the right secular trend—even if timing is off—tends to generate better returns if you can hold long enough.
- •JCal recounts passing on early Twitter because he over‑analyzed its shortcomings as a publishing medium, missing the network-effect upside.
- •Sacks notes PayPal’s founders succeeded partly because they weren’t ‘payments experts’ constrained by conventional wisdom.
- •Advice: startups must go all‑in on their best idea, not hedge across multiple mediocre ones; big companies, by contrast, can carry a portfolio of failures.
- 2:14:00 – 2:34:00
Google’s Gemini AI Overviews, Perplexity’s Squeeze, and Monopolies’ Second Chances
The crew dissects Google’s Gemini-powered AI Overviews, which effectively copy Perplexity’s answer-plus-citation model inside Google Search. They argue Google will likely grow revenues with more searches and better monetization, even as publishers suffer, and note how monopolies like Google and Microsoft can miss entire waves yet still win by fast-following innovators.
- •AI Overviews answer queries (e.g., ‘how to clean a fabric sofa’) with a synthesized guide and citations, plus ad slots—essentially Perplexity inside Google.
- •Friedberg expects fewer bounce‑backs, more on‑page engagement, and, in some verticals, higher revenue per query as Google internalizes more of the funnel.
- •Publishers whose content underpins these answers may lose traffic and pursue legal action; a rights marketplace for AI training is already being built by a founder Sacks knows.
- •Sacks emphasizes that Google and Microsoft repeatedly missed big shifts (LLMs, mobile) yet survived because their monopolies let them copy and recover.
- •Chamath updates his view, suggesting reports of Google’s ‘death’ in AI were premature, especially given call‑option businesses like Isomorphic Labs and Waymo.
- 2:34:00
Wrap-Up, All-In Summit Plugs, and March to One Million Subscribers
The episode closes with a quick victory lap on reaching 500K YouTube subscribers and plans for a live Q&A at one million. They plug the All‑In Summit, social channels, hiring a researcher, and JCal’s Athena investment, then sign off with their usual mix of inside jokes and banter.
- •Announcement: 500K YouTube subscribers reached; a live Q&A stream is promised as a reward.
- •All‑In Summit dates and application info, plus scholarship mention.
- •Plugs for each host’s side projects: Chamath’s Substack, Ohalo hiring, Sacks’ American Moment speech, JCal’s This Week in Startups and Athena discount link.
- •Teasing a future Vegas party at one million subscribers and jokes about their personal brands (Chamath’s unbuttoning, Freeberg’s AI fields, Sacks’ blazer, JCal as morning zoo DJ).
