• Chinese AI lab DeepSeek launched the DeepSeek-R1 model, rivaling OpenAI in math reasoning and code generation.

  • The model is (in part?*)open-sourced for global research use.

  • Requires way less computing power than competitors like Meta.

  • Competes with OpenAI in critical areas such as mathematical reasoning, code generation, and cost efficiency

  • Overcame U.S. chip export restrictions through optimized architecture.

  • Big Tech are sore loosers

*DeepSeek employs a dual licensing structure for its models. The codebase for DeepSeek-Coder-V2 is released under the MIT License, which allows unrestricted use, modification, and distribution. However, the pre-trained models are governed by the DeepSeek License Agreement, permitting research and commercial use with specific restrictions to prevent harmful applications. While DeepSeek’s models are open in many aspects, some argue they do not fully meet all criteria for being considered “open source” due to these licensing nuances

  • @GumusEnglish
    9•2 days ago

    I prefer to call these models “open-weights”. However, “open-source” is widely used and understood in this context. Not an intentional disinformation.