Why AI speeds ahead while robotics lags: Wooders thread, LeCun weighs in

On August 24, Sarah Wooders published a long thread answering why AI advances at lightning speed while robotics feels slow: the accelerators that unlocked mainstream AI are largely absent in robotics, though she also offers reasons for optimism. Yann LeCun added a theoretical explanation, and several practitioners contributed perspectives on timelines and industry conditions.

Confirmed

• Unlock 1: historical data. Wooders argues Transformers unlocked humanity's vast text archives—the main medium in which we store intelligence—while robotics has no equivalent: people rarely record mundane physical tasks like laundry, leaving robots without historical knowledge to learn from.
• Unlock 2: verifiable rewards (code). Code compiles, runs, and gives instant pass/fail signals, opening an entirely new interaction surface for AI over the past five years.
• Robotics breaks this loop. Even perfect code fails on clumsy or unmaintained hardware; real-world tasks lack clear intermediate rewards and offer infinite paths from A to B, making verification extremely hard—what she calls AI's biggest blind spot.
• Moravec's paradox persists. LeCun notes the real world is far more complex and messy than discrete symbol sequences, so the paradox (high-level reasoning is computationally cheap, low-level sensorimotor skills are not) still holds.
• Reasons for optimism. Synthetic data is becoming cheaper to build, and progress on non-verifiable rewards would benefit all of AI, not just robotics. Humanoid robots must also collect massive real-world data—why AI labs are investing in wearables—and overcome neural networks' unknown-unknowns and long-tail learning costs.
• Timeline view. @iandanforth frames it as a timeline offset: Transformer appeared in 2017 and embodied breakthroughs like RT-2 in 2023, suggesting robotics' exponential growth simply started years later.
• Industry context. @rasbt notes robotics progress is mostly hardware (e.g., a recent running record) with few practitioners and less buzz; @iScienceLuvr likens robotics to AI's GPT-3 era rather than its ChatGPT era; @gandamuml adds that talent historically fled the stagnant, heavily regulated robotics industry toward software, just as pragmatic people avoided AI until AlexNet and GPUs in 2012 proved it worked—hinting robotics could see a similar inflection point.

Why it matters

The thread turns a common complaint into structural analysis: data availability and verifiable rewards are AI's two engines, both missing in robotics. Breakthroughs in synthetic data or non-verifiable rewards would feed back into all of AI, and the timeline and talent-flow analogies suggest robotics could yet replicate AI-style explosive growth.

2026-08-24 ~ 2026-08-25 · 14 related posts

• [source] Why Is AI Moving at Lightspeed While Robotics Feels Slower? — sarahookr · 2026-08-24
• [source] Why AI moves fast but robotics lags: the data unlocking factor — sarahookr · 2026-08-24
• Robotics Lacks a Historical Store of Civilizational Knowledge — sarahookr · 2026-08-24
• Wooders: instantly verifiable code rewards were AI's second big unlock — sarahookr · 2026-08-24
• Real-World Tasks Lack Verifiable Rewards, AI's Blindspot — sarahookr · 2026-08-24
• Why robotics lags behind general AI: The challenge of rewards and verification — sarahookr · 2026-08-24
• Optimism for robotics: Synthetic data and solving non-verifiable problems — sarahookr · 2026-08-24
• Robotics faces data and reward challenges; synthetic data offers a key solution — sarahookr · 2026-08-24
• Why AI moves fast while robotics lags: barriers and talent flow — gandamu_ml · 2026-08-24
• Before AlexNet, pragmatic people avoided AI too — a robotics parallel — gandamu_ml · 2026-08-24
• Why does AI advance at lightspeed while robotics feels slower? — rasbt · 2026-08-24
• Robotics Lags Behind AI: Currently in its GPT-3 Era — iScienceLuvr · 2026-08-24
• Why Robotics Feels Slower Than AI Software — iandanforth · 2026-08-25
• [source] LeCun: Real World Complexity Keeps Moravec Paradox Relevant for Robotics — ylecun · 2026-08-25