media Hugging Face Forums · 9h ago · open_models

Hausa Healthcare LLM Preference & Safety Evaluation Dataset

from English

A computational linguist and Hausa NLP specialist has introduced a new dataset designed for evaluating Large Language Models in the context of healthcare and safety within the Hausa language.

The dataset was developed collaboratively by computational linguistics specialists and healthcare professionals, including doctors, nurses, and community healthcare workers. Healthcare experts were responsible for creating and reviewing the raw medical content, while linguistic specialists handled the Hausa language review, annotation, adaptation, and evaluation processes.

The author is seeking to connect with researchers and developers working on multilingual and low-resource healthcare AI, particularly those interested in Hausa-language evaluation and safety data.

Importance 1/3 Trust 1/3 Hugging Face Forums Safety & alignment Training data

Read original

arxiv arXiv cs.AI · 70d ago

Egy-DRiVeS uses Roop face-swapping to preserve pedestrian privacy and data usability

This work implements a five-stage pipeline for the Egy-DRiVeS dataset that protects pedestrian privacy through face swapping while maintaining essential facial attributes required for training autonomous vehicle models.

arxiv arXiv cs.CL · 71d ago · 1 view

PLURAL dataset improves LLM value alignment across 20 countries

Researchers introduce PLURAL, a large-scale preference dataset designed to help large language models reflect diverse global values rather than predominantly Western ones. The dataset is grounded in the Integrated Values Survey, which spans 92 countries, and contains approximately 500,000 synthetic preference triplets representing people from 20 diverse nations.

arxiv arXiv cs.CL · 73d ago · 3 views

Study profiles counter-misinformation ecosystem on Twitter

Researchers analyzed a large corpus of COVID-19 tweets using a domain-specific NLI model to classify 264,737 posts as supporting or opposing false claims. The study compared 23 user- and text-level features across these two groups to understand the dynamics of the counter-misinformation ecosystem.

media Hugging Face Forums · 74d ago · 2 views

Needpedia founder proposes human-annotated NVC dataset to address AI sycophancy

Anthony Brasher, founder of Needpedia, argues that the absence of publicly available Nonviolent Communication (NVC)-annotated training data is a critical gap in AI alignment research. He proposes creating a real, human-annotated corpus to help models understand the motivational structure beneath human language rather than merely simulating empathy.

arxiv arXiv cs.LG · 82d ago

FairBED: A Bayesian Experimental Design Approach to Gathering Fairer Data

The article introduces FairBED, a framework that modifies the data acquisition process itself to gather inherently fairer data, addressing biases present in existing datasets. It provides novel formulations for quantifying dataset fairness based on the principle that fair datasets should be uninformative about sensitive attributes.