Tag
A technology and human-rights researcher ran an experiment comparing three AI agents—Meta's Muse, Anthropic's Claude Cowork, and OpenAI's GPT 6.1 Sol—on a real-world task of filling gaps in the World Bank's Global Public Procurement Database in English (US) versus Farsi (Iran), revealing how language and context shape agentic performance and access to information.
Google advances AI for language support with new Gemini models for real-time translation and transcription, aiming to cover 1,000 languages through initiatives like the Universal Speech Model.
This paper presents Meddies-PII, a framework for generating a multilingual synthetic clinical dataset to train a model for personally identifiable information extraction, achieving state-of-the-art performance on benchmarks.
The paper introduces NepOOC, the first public Nepali-dominant multilingual benchmark for out-of-context misinformation detection, and evaluates multimodal architectures, finding that text-only models achieve strong performance.
This position paper argues that AI leaderboards are structurally ill-suited to serving the Global South due to lack of independent governance and conflict-of-interest policies, using India as a case study to highlight institutional barriers and propose solutions.
Nonprofit Current AI is building open, public AI infrastructure to serve underserved languages and communities, backed by $400M in funding from governments and foundations. It has launched Suno Sutra, an offline AI device supporting 22 Indian languages, and is funding open-source AI projects.
An experiment running the same research prompt about LENR and superconductivity through six AI systems in five languages reveals significant linguistic bias, with non-English queries surfacing information about real industrial commitments that English-only searches miss.
The paper introduces Macro, a preference alignment framework using DPO to improve the validity and minimality of self-generated counterfactual explanations across multiple languages.