Grok 4 Wows, The Bitter Lesson, Third Party, AI Browsers, SCOTUS backs POTUS on RIFs

Watch on YouTube ↗  |  July 11, 2025 at 22:13  |  1:30:50  |  All-In Podcast
Speakers
Chamath Palihapitiya — CEO, Social Capital
Travis Kalanick — Founder & CEO, Atoms
Keith Rabois — Managing Director, Khosla Ventures
Jason Calacanis — Angel Investor / Founder, LAUNCH

Summary

All-In hosts Jason Calacanis and Chamath Palihapitiya are joined by Travis Kalanick and Keith Rabois. Kalanick describes CloudKitchens' robotic food assembly and reported inbound interest in a Pony.ai and Uber autonomy venture, while the panel treats xAI's Grok 4 launch as confirmation of Rich Sutton's bitter lesson that general computation beats human-engineered knowledge. That framing drives critical views on Meta's Scale AI purchase, human-labeling startups, Google search and Apple's AI position, plus a debate over AI-native browsers from Perplexity and OpenAI. The episode closes with Elon Musk's proposed third party and the Supreme Court decision backing the administration on federal workforce reduction plans.

  • Travis Kalanick details CloudKitchens' robotic assembly line, which he says cuts kitchen labor from about 30-35% of revenue to 7-10%.
  • Kalanick confirms inbound interest in an autonomy venture tied to Pony.ai and Uber, but stresses no deal exists yet.
  • Chamath Palihapitiya frames Grok 4's benchmark lead as proof of Rich Sutton's bitter lesson: general computation beats human knowledge.
  • He argues Meta's roughly $15 billion purchase of 49% of Scale AI, and similar human-labeling bets, are now in question.
  • Keith Rabois says human-labeled data has a very short half-life, undermining Scale AI, Mercor and Surge.
  • Rabois calls Google search toast and says Apple has missed every AI window; Chamath calls browser-building poor capital allocation and urges Perplexity to replace Bloomberg.
  • The panel debates Elon Musk's proposed American Party, agreeing a few House seats are plausible but a Senate seat is not.
  • The Supreme Court backed the administration on planning federal workforce reductions, which Chamath calls an important curb on regulatory growth.
Ideas
Chamath Palihapitiya CEO, Social Capital 23:09
Meta's Scale AI bet backs losing approach
Chamath argues the bitter lesson calls into question the strategy of every lab that leaned on human knowledge, and singles out Meta: it spent about $15 billion for 49% of Scale AI, which he says is exactly a bet on human labeling. If general computation plus synthetic data make human-labeled data obsolete, that purchase and the strategy behind it lose their value, and Grok 4's results are the evidence that the shift is already happening.
Chamath Palihapitiya CEO, Social Capital 23:32
Camera-only FSD scaling beats Waymo's approach
Grok 4's benchmark lead validates Rich Sutton's bitter lesson: approaches that scale with raw computation beat approaches that depend on human knowledge and human labeling. Chamath says Elon Musk made the same architectural bet at Tesla years earlier, choosing camera-only FSD without LiDAR and betting that collecting billions of driving miles and applying general compute would reach autonomy faster than Waymo's more laborious and far more expensive engineered approach. He treats the Grok 4 result as evidence that Tesla's general-computation path to autonomy is the one that compounds.
Travis Kalanick Founder & CEO, Atoms 25:19
Hardware 5 should expand FSD capability
Travis argues Tesla is deliberately approximating human driving with cameras instead of LiDAR, on the logic that a human drives with two eyes and no spinning sensor, and that the missing piece has been onboard compute. He says Hardware 5 is coming to Tesla probably next year and that it will make a big difference in what FSD can do, making that compute upgrade the concrete event to watch on Tesla's autonomy timeline.
Keith Rabois Managing Director, Khosla Ventures 44:43
Apple keeps missing every AI window
Keith says Apple has missed every possible window on AI and continues to miss it, calling its performance on the most important technology breakthrough of the last seventy years absolutely miserable. He attributes this to CEO, cultural and infrastructure problems that are not an easy fix, and argues buying Perplexity would not help because there is no residual value there beyond some product taste. He concedes Apple is still alive and worth trillions only because it is vertically integrated, which is a reason it survives rather than a reason it wins in AI.
Keith Rabois Managing Director, Khosla Ventures 48:58
Google search is toast
Keith says Google search is toast and that Google will lose the search game to ChatGPT, which is already becoming the verb consumers use for AI. He argues Google owns the right pieces, Chrome for distribution and Gemini for model quality, and should be combining them into an agentic browser instead of defending search. His conclusion is that the best assets at Google now have nothing to do with search, and those other products are the only thing that can save the company.
Keith Rabois Managing Director, Khosla Ventures 55:34
Meta's AI spending missed the boat
Keith says Grok 4 shows that Mark Zuckerberg does not need to simply spend money to assemble a whole new AI team, because everything Meta has done in AI has also missed the boat. In his view the enormous sums Meta is now committing to AI talent are not buying it a competitive position, which makes the spend a capital-allocation problem rather than a fix.
Up Next

This All-In Podcast video, published July 11, 2025, features Chamath Palihapitiya, Travis Kalanick, Keith Rabois discussing META, TSLA, AAPL, GOOGL. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Chamath Palihapitiya, Travis Kalanick, Keith Rabois  · Tickers: META, TSLA, AAPL, GOOGL