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

Travis Kalanick says his food-robotics system is moving from prototype to customer deployment this quarter, cutting kitchen labor from roughly 30% of revenue to 7–10%. The 60-square-foot machine can assemble about 300 bowls an hour, portion ingredients precisely, bag orders, and place them in courier lockers.

Kalanick is also being approached about autonomous-vehicle projects, but says no deal exists yet. He framed autonomy as a shared logistics layer for moving both people and food, potentially linking his automated kitchens to driverless delivery.

Grok 4’s benchmark performance prompted the panel to question whether expensive human-labeled training data is becoming a rapidly depreciating asset. Palihapitiya emphasized compute-heavy learning and synthetic data, while Rabois estimated many labeling businesses may face only a one-to-three-year window as models increasingly label data themselves.

Kalanick cautioned that today’s mainstream LLMs still do not reliably invent genuinely new ideas; in his experience they cling hard to conventional knowledge. He thinks the breakthrough comes when models can generate hypotheses autonomously, but stressed that scientific claims still require physical experiments and verification.

Kalanick says consumer-software CEOs are already asking him how their businesses survive once agents replace direct app and webpage use. His expected end state is a concierge-like interface that simply handles tasks—such as booking travel—while many current consumer interfaces become invisible infrastructure.

The panel split sharply on Perplexity’s Comet browser: Rabois called it a necessary “Hail Mary,” while Palihapitiya argued Perplexity should attack Bloomberg instead. Palihapitiya said Bloomberg’s roughly $25,000 terminal has stale usability and that Perplexity’s financial-data work could target a far larger, defensible enterprise opportunity.

On Elon Musk’s proposed third party, the panel saw more leverage in a handful of congressional seats than in a presidential bid. Rabois stressed the difficulty of third-party Senate victories, while Palihapitiya pointed to 2023 FEC guidance expanding super-PAC ground operations as a mechanism Musk could potentially use.

Kalanick has quietly bought XG, the leading backgammon analysis engine, and plans to revive it with modern machine learning, larger compute and new training tools. He has also started competing himself, cashing in his first tournament and discovering attendees recognized him as XG’s owner rather than Uber’s founder.