DeepSeek, a company that shook up the AI world last month by releasing a new simulated reasoning model under an MIT license, is set to make its underlying code more accessible. The company plans to release five open-source repositories starting next week as part of its “Open Source Week”.
These daily releases will provide insight into the inner workings of DeepSeek’s online service, which has been battle-tested in production. The move aims to increase transparency and accelerate progress in the AI community.
In contrast, OpenAI’s ChatGPT models remain proprietary, with their inner workings opaque to outside users and researchers. This openness could help widen the gap between DeepSeek and OpenAI, but also provide wider access for end-users who are currently facing restrictions on DeepSeek’s mobile app due to privacy concerns.
The initial model release included “open weights” – a feature that allows users to fine-tune the model parameters with additional training data. This level of openness sets a precedent for DeepSeek’s code release and could be seen as a commitment to transparency and collaboration in the AI community.
Source: https://arstechnica.com/ai/2025/02/deepseek-goes-beyond-open-weights-ai-with-plans-for-source-code-release