Techyon

VibeLifeBench: Can Your Life Agent Be Proactive and Persistent in a Living World?

VibeLifeBench introduces a benchmark of 200 long-horizon, multi-week tasks across ten everyday-life domains to test LLM agents' proactivity and persistence in a dynamically changing simulated world. Agents must act witho…

→ View original source

Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation

The study applies reference-free post‑training to open large language models for multilingual machine translation, using Group Relative Policy Optimization with a reward that averages two reference‑free quality estimatio…

→ View original source

Cultivar: A Contrastive and Locale-Oriented Translation Benchmark for Investigating Contamination and Localisation Robustness

Researchers introduce Cultivar, a contrastive and locale-oriented translation benchmark derived from a localized subset of FLORES, designed to address contamination and localization challenges in multilingual translation…

→ View original source