On March 14, we submitted Hugging Face’s response to the White House Office of Science and Technology Policy’s request for information on the White House AI Motion Plan. We took this chance to (re-)assert the elemental role that open AI systems and open science play in enabling the technology to be more performant and efficient, broadly and reliably adopted, and meeting the best standards of security. This blog post provides a summary of our response, the complete text is obtainable here.
Context: Don’t Sleep on (Strongly) Open Models’ Capabilities
Open approaches to AI development are usually not only (typically) more transparent, adaptable, and scientifically sound, they’ve also consistently reproduced or surpassed the performance of widely-used API-only business offerings on many tasks; and are increasingly doing so on shorter timelines, with increased resource efficiency.
Our team’s recent OlympicCoder outperforming Claude 3.7 on complex coding tasks with 7B parameters and an open-source post-training recipe, or AI2’s fully open OLMo 2 models (with open training data) matching o1-mini performances, are two of essentially the most recent compelling examples.
These successes show that a strong AI strategy must leverage open and collaborative development to best drive performance, adoption, and security of the technology. We make three major recommendations on this direction.
Suggestion 1: Recognize Open Source and Open Science as Fundamental to AI Success
Essentially the most advanced AI systems thus far all stand on a powerful foundation of open research (attention mechanisms, transformer architectures, cheaper post-training algorithms) and open source software (PyTorch, Hugging Face libraries, supercomputer operating systems) — which shows the critical value of continued support for openness in sustaining further progress. Investment in systems that may freely be re-used and adapted has also been shown to have a strong economic impact multiplying effect, driving a significant percentage of nations’ GDP. As AI systems with open weights and training techniques turn into increasingly attractive options for developers by way of each performance and value, prioritizing public research infrastructure and broad access to compute, customizable models, and trusted open datasets — especially for smaller developers and researchers — shall be essential to the further technical and economic success of AI technology.
Suggestion 2: Prioritize Efficiency and Reliability to Unlock Broad Innovation
Addressing the resource constraints of organizations adopting and adapting AI technology shall be essential to supporting its diffusion and fostering innovation from adopters across your complete development chain. Smaller models (that will even be used on edge devices), techniques to reduce computational requirements at inference, and efforts to facilitate mid-scale training for organizations with modest to moderate computational resources all support the event of models that meet the particular needs of their use context, especially in high-risk settings reminiscent of healthcare where fully generalist models have proven unreliable.
More efficient and purpose-designed AI systems facilitate higher in-context evaluation, higher resource utilization, and enable organizations to construct technical capability in any respect stages of the AI development chain to be sure that all users can leverage the system that most closely fits their needs.
Suggestion 3: Secure AI through Open, Traceable, and Transparent Systems
Finally, if many years of data security and cybersecurity in open source software are any indication, open and transparent AI systems may have a fundamental role to play in securing AI development and deployment especially in essentially the most critical settings — with different levels of openness needed for various security requirements. Fully transparent models providing access to their training data and procedures can support essentially the most extensive safety certifications. Open infrastructure and open-source tooling implementing the latest training techniques can empower organizations to coach the models they need in fully controlled environments. Open-weight models that might be run in air-gapped environments could be a critical component in managing information risks. Prioritizing adoption of essentially the most transparent systems, supporting the event of the open resources outlined, and constructing capability to leverage them especially in critical settings of AI adoption are essential to enabling safer AI adoption.
Please confer with the full response for our more detailed recommendations!
