UPDATED NOVEMBER 22, 2026
UPDATED NOVEMBER 22, 2026

The Frontier

Your signal. Your price.

Include
Lookback
||
No Priors: Artificial Intelligence | Technology | Startups
  • · 3d ago

    Bob McGrew argues that superintelligence requires physical capacity as well as digital intelligence: AI has changed digital work, while everyday physical life looks little different from 10 years ago.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew rejects the claim that automating researchers creates runaway self-improvement. Researcher automation can deliver a major one-time gain, but progress eventually hits compute capacity and infrastructure limits.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew says experiments, not idea generation, drive scientific discovery. Better simulations can reduce lab work temporarily, but faster digital research will soon make physical experiments the bottleneck again.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew says American manufacturing loses on labor costs, while product designs change too often for bespoke automation to pay off. A new iPhone generation changes about 75% of its parts, leaving the rest suited to established automation.

    +2 more
    +2 more
  • · 3d ago

    Bob McGrew says conventional factory automation works best for stable products: a car line can justify years of setup, but rapidly changing electronics need automation that can be redesigned affordably.

    +2 more
    +1 more
  • · 3d ago

    Bob McGrew sees defense, consumer electronics, and data-center supply chains as strong markets for flexible production. In Ukraine, battlefield changes can prompt physical product-design changes within weeks.

    +2 more
    +2 more
  • · 3d ago

    Bob McGrew says manufacturing robots usually do not need humanoid bodies: one or two arms and a camera can cover many tasks. Existing arms can handle most assembly; inflexible, expensive automation is the bigger obstacle.

    +2 more
    +1 more
  • · 3d ago

    Bob McGrew distinguishes closed-world factory work from open-world tasks such as cleaning an unfamiliar bedroom. Factories can use perception and reinforcement learning to handle small variations without precisely calibrated, deterministic fixtures.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew says Arda is building software that captures manufacturing processes in a shared system, then uses AI to help engineers plan and reconfigure production lines. A first Anduril user cut an actuator virtual build from days to 90 minutes.

    +2 more
    +3 more
  • · 3d ago

    Bob McGrew favors problem-first software over training frontier models at Arda. Frontier models can interpret manufacturing processes and orchestrate specialized tools, while purpose-built algorithms handle tasks such as motion planning and CAD inspection.

    +3 more
    +2 more
  • · 3d ago

    Bob McGrew says robotics foundation models can learn useful world representations from broad data, including internet video, rather than relying mainly on scarce robot demonstrations. Fine-tuning for specific tasks remains an opening for robotics companies.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew says physical robots need fast control cycles, which large models may not meet. Distilled, smaller, and on-device models can improve speed and safety; open-source systems may benefit from data and research produced by major labs.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew applies Palantir's forward-deployed engineering approach at Arda: embed engineers with a customer, solve a concrete problem, and turn the lessons into reusable product. A sponsor matters, but so does a hands-on user willing to work through early failures.

    +2 more
    +3 more
  • · 3d ago

    Bob McGrew traces his shift from skepticism about neural networks to OpenAI through early robotics work. The team spent two years teaching a Shadow Hand to solve a Rubik's Cube; the project convinced him that robotic hands were exceptionally difficult to make useful.

    +3 more
    +2 more
  • · 3d ago

    Bob McGrew expects AI to connect cheap intelligence with perception and action: drones, sidewalk robots, and machines built for different environments will let systems do physical work beyond the data center.

    +3 more
    +1 more
  • · 3d ago

    Bob McGrew sees machine tools and factories that build other factories as an early form of physical self-replication. Automating factory design and construction could support the infrastructure needed to expand data centers and, eventually, machine activity into space.

    +2 more
    +1 more
  • · 3d ago

    Bob McGrew argues that self-replicating probes could spread civilization through space, and that their speed may explain why we have not detected other civilizations before they arrive. He also allows that humanity may be first or simply too distant to meet.

    +3 more
    +1 more
About The Frontier
17 results
End of 7-day results — 17 results