10 samples expanded to 242 languages, Adaption Labs aims to address AI multilingual shortcomings at the data level

ME News Report, April 15 (UTC+8), according to Beating Monitoring, AI data platform Adaption Labs released a new feature for Adaptive Data called “Expand Your World,” which can generate up to 2,420 high-quality training samples covering 242 languages and regional variants from as few as 10 samples in a single language, without additional annotation processes or data pipelines. This feature is now available to all Adaptive Data users.
Multilingual coverage is one of the main shortcomings of AI training data. Most datasets focus on a few high-resource languages, and models’ ability to handle minority languages and regional dialects is significantly weaker, making fine-tuning later difficult to fully compensate.
Adaption Labs’ approach is to front-load language coverage at the data level, addressing distribution bias during the data generation phase.
Adaption Labs was co-founded by Sara Hooker, former Vice President of Research at Cohere, and Sudip Roy, former Google AI Infrastructure Engineer. In February this year, it raised a $50 million seed round led by Emergence Capital, with a valuation of $1 billion.
The company’s core bet is to replace brute-force scaling with an efficient adaptive system, enabling models to continuously learn and evolve.
(Source: BlockBeats)