Votee AI, a Hong Kong-based startup, is developing large language models specifically for Cantonese, a language spoken by over 80 million people but underserved by current AI technology. By retraining open-weight models from developers like Meta and Alibaba on extensive Cantonese datasets, Votee aims to make AI more accessible and useful for local education, healthcare, and communication. Their approach involves a combination of online scraping, community data, and synthetic data generation, resulting in models that, while smaller than frontier models, can effectively understand and reason in Cantonese at a significantly lower training cost. AI
IMPACT Addresses the critical need for AI models in non-dominant languages, potentially unlocking AI's utility for millions globally.
RANK_REASON Focuses on a specific language niche within AI development, highlighting a gap in current LLM capabilities and a startup's efforts to address it. [lever_c_demoted from significant: ic=1 ai=1.0]
- Alibaba
- Anthropic
- Cantonese
- DeepSeek
- English
- Llama
- Meta
- Moonshot AI
- OpenAI
- Pak-Sun Ting
- Qwen
- Mandarin Chinese
- Votee AI
- Z.ai
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →