Using a model to produce training or test data in place of collecting real examples, generally to fill gaps where real data is scarce, restricted or expensive.

Offered as an unlimited supply of training data, which skips the part where generated data carries the assumptions of the model that produced it, and those assumptions are then trained straight back in.

• Unlimited Supply, Finite Source: Data generated by a model can only contain what that model already encodes.
• Circularity Unmentioned: Training on your own outputs narrows a model over time, and the pitch describes it as scaling.
• House-Cleaning Reality Check: Like photocopying the only photograph you own and describing it as a growing family album.

Love Island (Season 6): Casa Amor, a second villa of new arrivals introduced by producers for the specific purpose of generating material the existing villa was not producing on its own.

Using a model to produce training data instead of collecting real examples.

“Synthetic data generation gives us effectively unlimited training examples.”

“We generated 50,000 example clauses with a model, and they only contain patterns that model already knew.”