Verifier’s Law: Asymmetry of Verification Blog Post
Source: https://www.jasonwei.net/blog/asymmetry-of-verification-and-verifiers-law (via tweet) Author: Jason Wei Date: 2025-2026
Summary
Jason Wei’s blog post on asymmetry of verification and “Verifier’s Law” — the idea that verification asymmetry (easy to verify, hard to solve) is becoming foundational as RL finally works for general tasks. The tweet promotes the post as an important read for understanding why RL works now.
Key Claims
- Verification asymmetry is the key theoretical foundation for RL from code/math
- “Verifier’s Law”: as verification asymmetry increases for a domain, RL effectiveness scales
- This is why RL from code works (unit tests = cheap verification of expensive generation)
- Generalizes to: any domain with cheap verification can be RL-trained
Connection to Other Sources
Direct extension of Jason Wei’s Asymmetry paper. The blog post provides more accessible framing of the formal result.
Concepts
- RL Infrastructure — verification asymmetry as RL prerequisite
- Reward Hacking — what makes a good vs gameable verifier