Synthetic data & distillation
Model-generated training data as the real corpus runs dry
As high-quality human web text is exhausted, synthetic data and distillation (teacher-model generation, self-play, RL-from-verifiers) became the dominant frontier scaling lever in 2025 — central to reasoning models. It also fueled the DeepSeek distillation controversy and lab anti-distillation terms.
Driver
Human data exhaustion
Method
Distillation, RLVR
How it fits the stack
Synthetic data & distillation with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
Synthetic data & distillation in the AI stack. Synthetic data & distillation with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 7 entities, 7 relationships
- Anthropic —uses→ Synthetic data & distillation
- DeepSeek —uses→ Synthetic data & distillation
- DeepSeek-V3 / R1 —uses→ Synthetic data & distillation
- GPT-5 / o-series —used by→ Synthetic data & distillation
- GPT-5 / o-series —uses→ Synthetic data & distillation
- OpenAI —uses→ Synthetic data & distillation
- Synthetic data & distillation —competes with→ Training Data & Web Corpus
Feeds ↓ · 6
Context — capital, rivals, policy · · 1