Codex in Chrome 🤖, inside Chinese labs 🇨🇳, improving token efficiency 🛠️

AI model commoditization

AGI is not the ultimate scarce resource Silicon Valley claims; intelligence is commoditizing like compute and bandwidth, meaning model superiority alone won't create durable competitive advantages.

RL data quality control

Most vendors selling RL data into frontier labs are failing multiple quality control gates simultaneously, and those who don't raise their QC bar will run into serious problems this year.

Chinese AI Labs Prioritize Non-Flashy Improvements Over Individual Ideas

Chinese AI scientists are culturally more willing to do unglamorous, incremental model improvement work rather than championing personal ideas, resulting in less gamification and more flexibility in adopting modern techniques.

Contradicts the Western narrative that innovation requires individual ownership and competitive internal tribes; Chinese labs reportedly treat peers with respect and deprioritize the business side of AI.

Real AI Winners Will Own Customer Relationships and Data, Not the Best Models

As AI models commoditize like compute and bandwidth, the companies with superior customer relationships and proprietary data—not model quality—will capture lasting value.

Challenges Silicon Valley's AGI-as-scarce-resource narrative by arguing intelligence is being commoditized and competitive moats will come from distribution and data, not model capability.