← The AI Systems Map
L0 · LabsDatasetGlobal

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.

usesusesusesused byusesusescompetes withSynthetic data &distillationLabsAnthropicDeepSeekDeepSeek-V3 / R1GPT-5 / o-seriesOpenAITraining Data & WebCorpus
Synthetic data & distillationFeeds ↓Related

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
  • AnthropicusesSynthetic data & distillation
  • DeepSeekusesSynthetic data & distillation
  • DeepSeek-V3 / R1usesSynthetic data & distillation
  • GPT-5 / o-seriesused bySynthetic data & distillation
  • GPT-5 / o-seriesusesSynthetic data & distillation
  • OpenAIusesSynthetic data & distillation
  • Synthetic data & distillationcompetes withTraining Data & Web Corpus