What if open-source AI focused less on giant models and more on reusable capabilities?

The open-source AI community could shift its focus from developing large general-purpose models to creating reusable, domain-specific capabilities. These specialized models, for areas like biology, Python, or accounting, could be distilled from existing open models. Developers could then combine these capabilities to build local tools, such as an accounting assistant using a small model and OCR. This approach, similar to Linux's growth through shared components, could make domain capabilities the fundamental unit of contribution in open-source AI.