Key Milestones
101
Claude 3 IQ Score (2024)
133
o3 IQ Score (2025)
10X
Compute Scaling (2020-2025)
1000X
Training Data Growth
The Five Scaling Laws
Compute Scaling Law
Performance scales predictably with compute power (FLOPs). Doubling compute typically improves performance by a predictable margin.
Data Scaling Law
More training data leads to better performance, following power-law relationships with diminishing returns.
Model Size Scaling
Larger models (more parameters) consistently outperform smaller ones when properly trained.
Inference Scaling
Test-time compute scaling allows models to "think" longer for better performance on complex tasks.
Emergent Abilities
New capabilities emerge unpredictably at certain scale thresholds, leading to breakthrough performances.
The Path Forward
The AI revolution is accelerating through predictable scaling laws. As we approach 2025-2030, we can expect:
Superhuman Intelligence
AI systems exceeding human experts in specialized domains
Global Impact
Transformative effects on science, technology, and society
Continuous Evolution
Rapid iteration and improvement in AI capabilities