Adaption Launches Invent a Dataset: Generate Training Data from Task Descriptions

Adaption Launches Invent a Dataset

Adaption has unveiled a new tool called Invent a Dataset, designed to streamline the generation of training data for AI models. Unlike conventional synthetic data tools that only produce samples, Invent a Dataset also handles the design of the dataset structure. Users simply describe the task they want the model to learn — such as "summarizing business documents according to company standards" or "categorizing customer service requests" — and the tool automatically determines the data schema, generates aligned samples, and outputs them ready for training. No raw data, predefined field formats, or detailed annotation rules are required. The generated data can be directly fed into Adaption's AutoScientist system, which iteratively adjusts training parameters, retrains and evaluates the model until performance plateaus. This enables a complete pipeline from a single task description to a fully trained model.