Adaption Labs launches dataset generation from task descriptions
Adaption Labs has released Invent a Dataset, a feature that creates structured, training-ready data from a description of the behavior a model should learn. The process does not require a seed corpus, predefined schema, or labeling guide. It is available in the Adaption app and through a Python SDK and REST API. Generation runs on the company’s hosted platform and uses credits; no self-hosted option is documented.
A call to datasets.invent starts an asynchronous job, which can be monitored through datasets.get. Users select one or more current domain codes, choose instruction_dataset for prompt-completion pairs or preference_pairs for chosen and rejected completions, and can guide the content with a prompt of up to 10,000 characters. Results can be downloaded as JSONL, JSON, CSV, or Parquet. Estimation and idempotency options are also provided, while row limits depend on the plan.
Language expansion can translate rows or localize them for country-language combinations. The resulting dataset ID can be passed to autoscientist.create. Adaption says AutoScientist, launched in May 2026, outperformed training configured by its research staff by an average of 35% in internal evaluations across eight verticals, with win rates rising from 48% to 64%.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.