Updates from Providers!

In continuation of our research, (see peer-reviewed article), we emailed the providers to get their perspectives and to share our analysis. Of these, we received responses from Swiss AI Initiative, the provider for Apertus, and from HuggingFace, the provider for SmolLM. Both providers were extremely positive about the work and its necessity and positioning within the ecosystem. Below we share some our thoughts regarding this engagement.

Swiss AI Initiative / Apertus

We found the Swiss AI Initiative to be very forthcoming and engaging with us with good faith, both in terms of understanding the analysis as well as expressing their desire to improve their public summary using our assessment. Based on the public summary and the engagement as well as the Swiss AI Initiative's conduct, we consider their summary to be a role model for the implementation of the AI Act and its transparency obligations. We also consider their summary's visibility to be an important aspect for the open-source ecosystem in terms of AI Act compliance, and believe their contributions to be influential in further development of the Regulation and its implementation.

HuggingFace / SmolLM

HuggingFace also responded in what we felt as good-faith discussions regarding various aspects of our data quality assessment framework. In particular, it helped surface potential issues with alternative interpretations on specific parts of the template. We have incorporated these into our ongoing refinement considerations.

Conclusion

These early engagement from providers helped us understand their perspectives, especially where certain things need to be additionally clarified, and where our assessment methodology needs to evolve to guide this process. It also showcased that there are providers with good quality summaries and they such providers engage to uphold both the spirit and the letter of the law. We believe this has set the bar for the public summaries yet to come.