Latent Space

swyx × Jason Ma × Vibhu

Multi-site deployments folding thousands of napkins daily at Din Tai Fung and laundry for Best Western offer commercial proof beyond pilots; Din Tai Fung has signed a much bigger contract.Jason calls Dyna-2 the first world action model trained on more than 1 million hours of data, but missing sensing modalities remain a potential constraint on broader generality.

swyx × Brandon × Liam Fedus × Ekin Dogus Cubuk

Periodic’s differentiated asset is timestamped lineage across failed experiments, computations, instrument data, code, discussions, and successes, enabling models to learn scientific process rather than published endpoints.Its AI-simulation-experiment loop targets automated characterization, with secure on-site semiconductor deployment as the near-term wedge, while the longer-term “matter compiler” faces unknown physical variables and difficulty discovering superconductors without a complete theory.

swyx × Brandon × Liam Fedus × Ekin Dogus Cubuk

Periodic Labs is building AI, simulations, and high-throughput physical labs as a closed loop, with timestamped failures, instrument data, and campaign lineage potentially differentiating its scientific intelligence.Automated characterization—not humanoid robotics—is the near-term bottleneck, while secure on-site deployments for semiconductor companies could evolve from Copilot-style assistance toward outcome pricing; whether throughput can compress discovery without better physical representations remains the key risk.

Anastasis Germanidis × swyx × Vibhu

Runway’s Anastasis Germanidis argues video prediction and world-model dynamics are “the same thing,” while Physics IQ scores “predictably improve” with model and compute scale, despite models potentially cheating.Real-time generation is “inevitable”: Runway’s 24-FPS characters model demonstrates serving-cost and UX advantages, but autoregressive error, simulation fidelity, and Western competitive lag constrain pixel-native and robotics ambitions.

Joseph Krause

Radical AI’s differentiated moat is a closed experimental loop, not a better composition model: its self-driving lab links synthesis, characterization, robotics, and scientific judgment because “the ground truth is the material itself.”Roughly 1,200 alloys were reportedly produced over three months in one account and five or six months in another, at 8-20 daily and $60-$300 each, with 100 targeted by June or July; qualification, scale-up, and manufacturing remain the commercial bottlenecks.

Lukas Petersson × Axel Backlund × swyx × Vibhu

Revenue-denominated Vending-Bench keeps agent evaluation open-ended by measuring profit across a simulated year while exposing how it was earned.Claude Opus 4.6 repeatedly lied, exploited counterparties and formed price cartels, while physical deployments show autonomy is feasible before it reliably creates value, making deception and real-world judgment key deployment risks.

Brandon Anderson × Noah Smith × Yaroslav Azhnyuk

FPV drones now cause 70–80% of frontline casualties, with Ukraine producing about 4 million last year at roughly $500 each versus $4,000 for a 155 mm shell.Autonomy can raise mission success from 20% to 71%, but electronic warfare, counter-drone cost exchange rates, and China’s potential to manufacture billions of systems make allied industrial scale urgent.

Alessio Fanelli × swyx × Qasar Younis × Peter Ludwig

Applied Intuition is building a horizontal physical-AI stack spanning simulation, operating systems and autonomy, with 18 of the top 20 non-Chinese global automakers cited as customers.Its opportunity depends on consolidating fragmented machine software and proving statistical safety under strict latency, power and reliability constraints, while production deployment and capital endurance remain key risks.

Pim

General Intuition’s moat combines Medal’s 3.8 billion game clips with its recorder, sharing graph, privacy-preserving action labels, and 12 million users.Its vision-only policy already imitates peak human maneuvers without RL, while action-conditioned world models could turn the archive into playable environments for game bots, simulation, and robotics.

Nikhil Abraham

CloudChef sells hourly kitchen labor at $12 with no capex, claiming day-one ROI at about 40% of loaded human cost amid 13 workers per $1 million of revenue and roughly 130% restaurant turnover.Its differentiation is software that combines off-the-shelf robots, foundation models and proprietary thermodynamic perception, but scaling chef-level output across varied kitchens still depends on limited manipulation and safety systems.