Physical AI Hits Its "GPT-2 Era" as Robotics Data Bottleneck Bites

Skynet Chance (-0.03%): The article documents that embodied AI remains unreliable and data-starved, meaning general-purpose robots capable of autonomous real-world action are further from existing than investor enthusiasm suggests, slightly lowering near-term loss-of-control risk. However, growing convergence of AV world models and humanoid embodiment infrastructure keeps a long-run pathway toward capable physical agents open.

Skynet Date (+0 days): The reported data crisis, halved Unitree valuation, and admission that manipulation is "like self-driving five years ago" suggest embodied autonomy arrives later than hype implied, modestly pushing back the timeline for physically capable autonomous systems.

AGI Progress (+0.01%): Shared simulation, world-model, and ML-ops infrastructure spilling over from autonomous vehicles into humanoid robotics represents real if incremental progress on grounding AI in the physical world, a component many consider necessary for general intelligence. The candid GPT-2 analogy also frames a concrete, scaling-based path forward rather than a dead end.

AGI Date (+0 days): Billions in investment, new robotics labs at Wayve and Uber, and tooling like Foxglove's Cosmos-based data search modestly accelerate embodied AI iteration speed, though the acknowledged data and reliability bottlenecks temper how much the overall AGI timeline moves.