Research
Science
Papers and methods from the aixplain research team. Peer-reviewed, reproducible, and available for citation.
Showing 2 of 2
Science paperSafety2025
Safety Evaluation of Multilingual LLMs in Enterprise Deployments
Systematic evaluation of safety properties in multilingual large language models across Arabic, English, French, and Spanish. Identifies cultural alignment gaps and proposes mitigation strategies for enterprise use.
Science paperMultilingual2025
AlphaLoss: Adaptive Loss Functions for Multilingual Model Fine-Tuning
A novel adaptive loss function that improves fine-tuning stability across multilingual models, with particular gains for low-resource Arabic dialects. Demonstrates consistent improvements over standard cross-entropy baselines.
Need a custom evaluation?
We run independent benchmark programs for enterprise and government clients.