Sabitlenmiş Tweet
Unpopular opinion: General purpose robots will arrive faster than Level 5 self-driving cars!
Here is how we think the next 5 years play out:
➡️ Phase 1: The Narrow Era (2025-2027)
- Robots remain brittle. They require precise models of the world. The economics don't make sense yet. A human is cheaper, stronger, and smarter than a Unitree G1. This phase is purely for establishing baselines and gathering edge cases.
➡️ Phase 2: The Data Flywheel (2026-2028)
- We are starting from zero data. To solve general intelligence, we need volume. The goal of the first 100,000 units is not to be productive; it's to fail.
Deploy -> Fail -> RLHF -> Update.
Once the sim-to-real loop tightens, the exponential curve begins.
➡️ Phase 3: The Software-Defined Moment (2028+)
- Asking the robot to "Make me a sandwich" won't require coding; it will require a text prompt. When a robot can learn a task in 5 minutes via demonstration, the ROI becomes undeniable.
This is Zero-Shot Generalization.
Robotics will scale faster than self-driving cars for one reason: The Cost of Failure.
Self-driving car failure = Fatality.
Humanoid failure = Dropped laundry.
Hardware costs are already collapsing (Wright’s Law).
The data flywheel is starting to spin. We are leaving the era of Tele-operation and entering the era of Self-Improving Machines.
English
