Chinese AI firm releases DeepSeek V3, a new leader in open-source AI models

DeepSeek V3 demonstrates that high-performance AI is achievable with significantly lower costs through efficient architectural design. By leveraging a mixture of experts, it activates only necessary components, drastically reducing hardware expenses during both training and inference. This challenges the assumption that massive budgets are required for top-tier large language models. The model outperforms major closed and open-source competitors on most benchmarks, proving that open-source alternatives can rival proprietary systems like GPT-4o. Achieving this with a fraction of the typical industry training costs highlights the potential for more accessible and sustainable AI development. This is relevant to open data and open source because it accelerates transparency and innovation. By releasing superior code and weights under an accessible license, DeepSeek enables researchers and developers to build upon a strong, affordable foundation. This promotes a more competitive and collaborative ecosystem, reducing reliance on expensive, closed proprietary technologies and fostering broader technological advancement.

Source: thehindu.com
Published on 2024-12-28