Locai Labs Announces the Launch of NVIDIA Nemotron-Personas for the UK
1 million privacy-preserving synthetic personas grounded in demographic, geographic, and socio-economic distributions of the UK population.
Locai Labs Limited, the sovereign AI computer company based in London, today announced, in collaboration with NVIDIA, Nemotron-Personas-UK, an open-source dataset of 1 million privacy-preserving synthetic personas grounded in demographic, geographic, and socio-economic distributions of the UK population. Nemotron-Personas-UK enables developers to build and test UK sovereign AI systems that more accurately represent the UK’s rich diversity of people across regions, cultures, languages and demographics and hence reduce bias in AI models without using any personal or private data.
Example use cases for Nemotron-Personas-UK include:
● Training model personalisation engines that reflect the UK’s rich demographic and cultural variety
● Evaluating models in domain-specific applications, such as healthcare, law and finance
● Fine-tuning models on culturally grounded data
● Evaluating model bias across rural and urban populations, age groups and education levels, to check that systems perform consistently across the UK
● Simulating representative user interactions to evaluate and train conversational AI agents across realistic populations and languages
This dataset was developed on a hybrid set of 18 NVIDIA GPUs hosted on a combination of cloud and local compute, using Locai Labs’ own AI computer, the Locai 1. The Nemotron-Personas-UK dataset is fully open source, with a Creative Commons Attribution 4.0 (CC BY 4.0) licence, and will join the Nemotron Personas Collection, which comprises a growing set of countries including France, Japan, Brazil and the USA, with Germany also joining during GTC Berlin.
James Drayson CEO of Locai Labs said: “As Locai Labs is a London-based AI lab with a track record in developing LLMs for the UK, we were delighted to be chosen by NVIDIA to develop Nemotron-Personas-UK. We hope that this new open-source dataset will enable the development of a wide range of UK sovereign AI systems, properly aligned with the needs and diversity of the UK.”
Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI
Creating the dataset presented some unique challenges due to the differences across the four countries of the UK including their different census systems, different educational qualifications, and the variety and geographic distribution of languages spoken. Care was taken to factor Celtic language proficiency into UK persona generation given the likely use of the dataset in creating AI models that are proficient in the languages of the UK. Although the Nemotron-Personas-UK dataset is aligned to real demographic and occupational distributions of the UK populations no real person is represented in it.
The use of synthetic (i.e. computer generated) data has become a crucial part of training modern LLMs. Generating synthetic data at scale is challenging as it often suffers from a lack of diversity, which can lead to AI models being biased, reflecting the data used to train them rather than reflecting the diversity of their actual users. Seeding synthetic data generation with realistic personas that are representative of the population they are intended to serve is a powerful way of preserving diversity and reducing bias in model development.
The Nemotron-Personas-UK dataset contains:
● 1M personas (comprised of 6M persona fields & 18M contextual fields)
● 0.59B tokens (0.23B persona tokens)
● 372 local authorities across England, Wales, Scotland and Northern Ireland
● Records by nation, in proportion to the adult population: England & Wales 889,311 (88.9%); Scotland 83,163 (8.3%); Northern Ireland 27,526 (2.8%)
● 8,999 unique first names and 23,807 unique surnames
Also Read: AI and The Future of Work: Artificial Intelligence Is Expanding Organizational Intelligence Beyond Human Limits
[To share your insights with us, please write to psen@itechseries.com ]
Comments are closed.