Scaling Brilliance
Core philosophical framework of axiom-math and carina-hong describing how formal-verification enables both scaling and compounding of mathematical insights. The concept has two complementary dimensions that together enable exponential knowledge growth.
Two Dimensions
Scaling Dimension
Definition: Enabling more people to benefit from mathematical insights
- Formal proofs communicate intuitions reliably across individuals
- Verified results can be trusted and built upon by others
- Quality of formal proofs approaches human-level regardless of generating system
- High-quality training corpus grows through verified outputs
Compounding Dimension (compounding-brilliance)
Definition: Mathematical insights building upon each other over time
- Verified proofs provide solid foundations for further development
- Future inference and training can reliably build on proven results
- Avoids the degradation that occurs with informal, unverified reasoning
- Creates exponentially growing value from proven mathematical knowledge
Historical Analogy: Ramanujan
Hong uses srinivasa-ramanujan as the paradigmatic example. When G.H. Hardy persuaded Ramanujan to formalize his intuitive mathematical insights:
- Personal improvement: Formalization forced articulation of details, opening new lines of thinking
- Scaling effect: Others could understand, learn from, and validate his work
- Compounding effect: Future mathematicians could build reliably on his proven foundations
Technical Implementation
In AI systems, Scaling Brilliance manifests through:
- Reinforcement Learning with Verification: Using formal proof correctness as training signal
- Verified knowledge bases: Accumulating mathematically sound results over time
- Sample efficiency: Stronger verification signals vs statistical approaches (GRPO, RLHF)
- Maximum performance: Higher ceiling than informal reasoning approaches
Philosophical Foundation
The concept represents Axiom's core thesis that verification isn't about "fixing lousiness" but about amplifying and preserving brilliance. As Hong states: "Verification to me is about scaling brilliance, compounding brilliance."
This distinguishes their approach from traditional AI safety or correctness concerns, instead positioning verification as a fundamental enabler of intellectual progress.