Synthetic Data
Generating training data with models: privacy, preference data, weak supervision and model collapse.
- 1
The Looming Digital Ice Age: AI Eating Itself?
Is AI eating itself? Explore the "model collapse" and the "Hapsburg AI problem" before our digital world speaks only gibberish.
- 2
Can Fictional Twins Save AI From Running Out of Internet?
As high-quality human data runs dry, synthetic data is becoming the new gold standard for training the next generation of AI models.
- 3
Substitution Anonymization: Privacy Without Utility Loss
How to generate realistic synthetic voice notes and calendar data with zero PII exposure risk.
- 4
AI Training Itself: Student, Teacher, and Grader
Can models generate their own training data and judge their own outputs? The promise and pitfalls of fully AI-led pipelines.
- 5
Training a Model When You Don't Know the Labels Yet
Daniel wants to extract adapter specs from his shelf — but he can't name the labels until he's already annotating.
- 6
How to Build an AI Preference Dataset
Pairwise comparison, forced choice, or just "this one or that one" — the engineering behind collecting human preferences on AI outputs.
- 7
Who Checks the Labels? Weak vs Strong Supervision
Human labels, rule-based labels, or a teacher model writing the exam — what actually makes supervision "strong," and who verifies it?
- 8
Which AI Models Actually Generate Good Training Data?
Open weights vs closed APIs for synthetic data, and why IFEval scores don't tell you which model will hold up.
- 9
Small Models, Big Questions: When Cheap Fables Get Expensive
Three million fables for pennies — but can a small model tell a good example from a plausible one?
- 10
Synthetic Logs Without Leaking Your Clients
You can rewrite your logs with a cheap model — but can you keep the task boundaries while losing the client names?