Base Labs launches an open-weight AI safety partnership with Hugging Face (techcrunch.com)

Baseten has announced the launch of a safety infrastructure standard through its Base Labs research arm, collaborating with Hugging Face and Goodfire AI. This initiative aims to develop safety evaluation and monitoring systems for open-weight AI models, which have been increasingly subject to risks due to a technique known as abliterating, where crucial safety measures can be stripped away. With Hugging Face hosting over 6,000 models at risk, this partnership is a timely response to growing safety concerns within the AI community. The significance of this initiative lies in its commitment to integrate safety protocols directly into the training and deployment of open models, rather than applying them retroactively. By promoting transparency and accountability, Baseten and its partners believe that open-source frameworks can lead to safer AI development. Goodfire AI's expertise in model interpretability will likely play a crucial role in this effort. Furthermore, Baseten is encouraging contributions from the broader developer community to foster an ecosystem of accessible and safe open models, highlighting a collaborative approach to addressing AI safety challenges.