black box distillation
Distillation using only text outputs from an API-accessible model without access to internal weights or probability distributions; weaker signal than white box but accessible to external parties through public APIs
As defined in Model Distillation: How to Steal a Billion-Dollar AI
Where it comes up · 1
Most important claims · 4 of 4Search all claims for “black box distillation” →
- Black box distillation against models you do not own carries real legal risk; OpenAI, Anthropic, and Google all have terms of service that explicitly prohibit using their outputs to train competing models.Model Distillation: How to Steal a Billion-Dollar AI
- White box distillation—accessing a model's internal probability distributions—requires owning the model; outside laboratories like OpenAI or Google, users only have access to public APIs that return text outputs, forcing reliance on weaker black box distillation signals.Model Distillation: How to Steal a Billion-Dollar AI
- OpenAI and Microsoft accused DeepSeek of training R1 partly on ChatGPT's own outputs, which is textbook black box distillation and strictly prohibited under OpenAI's terms of use.Model Distillation: How to Steal a Billion-Dollar AI
- In black box distillation, only text outputs are visible; prompts are sent, answers are saved, and small models are trained on those outputs alone, which is a weaker signal than soft labels but repeated across 100,000 examples burns the teacher's habits into the student.Model Distillation: How to Steal a Billion-Dollar AI