PulseAugur
EN
LIVE 18:30:07

New SomaliBench benchmark reveals large refusal gaps in open-weight LLMs

By PulseAugur Editorial · [3 sources] ·

A new benchmark, SomaliBench v0, has been developed to evaluate the safety refusal capabilities of open-weight language models in Somali, a low-resource language. The study found significant gaps in refusal rates between English and Somali for models like Llama-3.1-8B-Instruct, Aya-23-8B, Qwen-2.5-7B-Instruct, and Gemma-2-9B-Instruct. For many models, non-refusal in Somali often resulted in unclear or incoherent outputs rather than direct harmful compliance. AI

IMPACT Highlights the need for more robust safety evaluations in low-resource languages, potentially influencing future model development and testing.

RANK_REASON The cluster describes a new academic benchmark and evaluation of existing models, fitting the research bucket.

Read on Hugging Face Daily Papers →

AI-generated summary · Google Gemini · from 3 sources. How we write summaries →

New SomaliBench benchmark reveals large refusal gaps in open-weight LLMs

COVERAGE [3]

  1. arXiv cs.AI TIER_1 English(EN) · Khalid Yusuf Dahir ·

    SomaliBench Eval: Measuring English-to-Somali Refusal Gaps in Open-Weight Language Models

    arXiv:2605.25420v1 Announce Type: cross Abstract: Large language model safety evaluation remains heavily English-centered, leaving low-resource languages under-measured even when models are deployed globally. We evaluate four open-weight instruction-tuned models on SomaliBench v0…

  2. Hugging Face Daily Papers TIER_1 English(EN) ·

    SomaliBench Eval: Measuring English-to-Somali Refusal Gaps in Open-Weight Language Models

    Large language model safety evaluation remains heavily English-centered, leaving low-resource languages under-measured even when models are deployed globally. We evaluate four open-weight instruction-tuned models on SomaliBench v0, a native-author-verified benchmark of 100 harmfu…

  3. arXiv cs.CL TIER_1 English(EN) · Khalid Yusuf Dahir ·

    SomaliBench Eval: Measuring English-to-Somali Refusal Gaps in Open-Weight Language Models

    Large language model safety evaluation remains heavily English-centered, leaving low-resource languages under-measured even when models are deployed globally. We evaluate four open-weight instruction-tuned models on SomaliBench v0, a native-author-verified benchmark of 100 harmfu…

RELATED ENTITIES

RELATED TOPICS