World PulseNow
Newsletter

MERLIN: Multi-Stage Curriculum Alignment for Multilingual Encoder-LLM Integration in Cross-Lingual Reasoning

arXiv — cs.CL•Wednesday, November 12, 2025 at 5:00:00 AM
PositiveArtificial Intelligence
MERLIN represents a significant leap in the integration of multilingual capabilities within AI, particularly for low-resource languages (LRLs) that have historically been underserved by existing models. By employing a two-stage model-stacking framework and a curriculum learning strategy, MERLIN not only enhances the accuracy of cross-lingual reasoning but also adapts a minimal set of DoRA weights for efficiency. The model's performance on the AfriMGSM benchmark showcases a 12.9 percentage point improvement over MindMerger, alongside consistent gains on MGSM and MSVAMP benchmarks. This progress is vital as it narrows the performance gap that has persisted in LRLs, which are often neglected in AI advancements, ensuring that more languages can benefit from sophisticated language processing technologies.
— via World Pulse Now AI Editorial System
Read Original

Was this article worth reading? Share it

Recommended Readings
I Let an LLM Write JavaScript Inside My AI Runtime. Here’s What Happened
DEV Community8 hours ago
I Let an LLM Write JavaScript Inside My AI Runtime. Here’s What Happened
PositiveArtificial Intelligence
The article discusses an experiment where an AI model was allowed to write JavaScript code within a self-hosted runtime called Contenox. The author reflects on a concept regarding tool usage in AI, suggesting that models should generate code to utilize tools instead of direct calls. This approach was tested by executing the generated JavaScript within the Contenox environment, aiming to enhance the efficiency of AI workflows.
Read full article
via DEV Community
From Fact to Judgment: Investigating the Impact of Task Framing on LLM Conviction in Dialogue Systems
arXiv — cs.CL2 days ago
From Fact to Judgment: Investigating the Impact of Task Framing on LLM Conviction in Dialogue Systems
NeutralArtificial Intelligence
The article investigates the impact of task framing on the conviction of large language models (LLMs) in dialogue systems. It explores how LLMs assess tasks requiring social judgment, contrasting their performance on factual queries with conversational judgment tasks. The study reveals that reframing a task can significantly alter an LLM's judgment, particularly under conversational pressure, highlighting the complexities of LLM decision-making in social contexts.
Read full article
via arXiv — cs.CL
"As Eastern Powers, I will veto." : An Investigation of Nation-level Bias of Large Language Models in International Relations
arXiv — cs.CL2 days ago
"As Eastern Powers, I will veto." : An Investigation of Nation-level Bias of Large Language Models in International Relations
NeutralArtificial Intelligence
This paper systematically examines nation-level biases exhibited by Large Language Models (LLMs) in International Relations. Utilizing historical records from the United Nations Security Council (UNSC), a bias evaluation framework with three distinct tests was developed. Results indicate that biases vary across models, with general patterns showing favorable biases toward Western nations and unfavorable biases toward Russia. The study highlights that LLM biases are multidimensional and context-dependent, suggesting that models with stronger reasoning abilities exhibit reduced bias.
Read full article
via arXiv — cs.CL