Baseten Unveils Safety Framework for Open-Weight AI Models in Collaboration with Hugging Face and Goodfire AI

On Wednesday, Baseten, in partnership with Hugging Face and Goodfire AI, launched a new safety infrastructure standard via its research arm, Base Labs. This initiative addresses safety concerns tied to open-weight models, which are models that can become hazardous when safeguards are removed through a process called abliteration. Hugging Face currently hosts over 6,000 such models.

Key Points:

• Base Labs aims to establish transparent standards for training and deploying open models, integrating safety directly rather than adding it later.
• Baseten emphasizes the advantages of openness for AI safety, promoting the visibility and transparency of model behaviors.
• The partnership with Goodfire aims to embed safety within open models; Goodfire focuses on making AI processes more interpretable.
• Baseten and Goodfire are both significantly funded; Baseten recently raised $1.5 billion, while Goodfire secured $150 million.
• Baseten is inviting contributions from the developer community to bolster this framework and ensure open models are safely accessible.

The announcement is part of a broader effort to create safer, more transparent AI environments as concerns about the potential misuse of AI continue to grow.